Expand Assistent to v0.3: Generate loop, Civitai Confirm, img2img/inpaint.

Server inventory and streaming chat, auto-apply/generate with Interrupt, Civitai search cards (Confirm-only download), plus Init/Mask wiring and inpaint_edit pack.

Co-authored-by: Cursor <cursoragent@cursor.com>

Assets/assistent.css

@@ -201,6 +201,11 @@
     max-width: 12rem;
 }
 
+.sa-header-right .sa-model-select {
+    min-width: 14rem;
+    max-width: 24rem;
+}
+
 .sa-check {
     flex-direction: row !important;
     align-items: center;
@@ -279,6 +284,55 @@
     margin-top: 0.45rem;
 }
 
+.sa-danger {
+    opacity: 0.85;
+}
+
+.sa-civitai-list {
+    display: flex;
+    flex-direction: column;
+    gap: 0.45rem;
+    margin-top: 0.55rem;
+}
+
+.sa-civitai-card {
+    padding: 0.5rem 0.6rem;
+    border-radius: 0.45rem;
+    border: 1px solid color-mix(in srgb, currentColor 22%, transparent);
+    background: color-mix(in srgb, currentColor 5%, transparent);
+    font-size: 0.88rem;
+}
+
+.sa-civitai-card.sa-installed {
+    opacity: 0.7;
+}
+
+.sa-civitai-title {
+    font-weight: 600;
+    margin-bottom: 0.2rem;
+}
+
+.sa-civitai-meta {
+    opacity: 0.8;
+    font-size: 0.82rem;
+    margin-bottom: 0.35rem;
+}
+
+.sa-civitai-actions {
+    display: flex;
+    flex-wrap: wrap;
+    gap: 0.35rem;
+}
+
+.sa-streaming {
+    opacity: 0.9;
+    font-style: italic;
+}
+
+.sa-interrupt-active {
+    border-color: color-mix(in srgb, #c44 55%, transparent) !important;
+}
+
 .sa-composer {
     border-top: 1px solid color-mix(in srgb, currentColor 18%, transparent);
     padding: 0.6rem 0.7rem 0.7rem;

Prompts/base_krea2.md

@@ -15,15 +15,18 @@ You are **Swarm Assistent**, a collaborative art director for **Krea 2** image g
 
 A JSON block named "Live SwarmUI context" is attached. Treat it as ground truth:
 
-- Use only LoRAs listed in `available_loras` (by exact `name`).
+- Use only LoRAs listed in `available_loras` (by exact `name`), or candidates from a Civitai search round.
 - Prefer listed `trigger_phrase` / `triggers` — **never invent** trigger words.
 - When enabling a LoRA, include its triggers in `prompt` if missing.
-- Respect current width/height/steps/cfg unless the user asks to change them or the pack is `fix_params`.
+- Respect current width/height/steps/cfg/seed/sigma_shift/sampler unless the user asks to change them or the pack is `fix_params`.
+- `wildcards` lists installed wildcard names (`__name__` syntax in prompts).
+- `prompt_image_count` > 0 means Prompt Images are attached — warn if they may dominate.
+- **Init / inpaint:** `has_init_image`, `has_mask_image`, `init_creativity` (aka denoise, 0–1), `mask_blur`, `mask_grow`. `has_vision_image` is the Assistent pane reference (can become Init/Mask).
 
 ## Output contract (mandatory)
 
 1. Write a short helpful reply in the user's language (RU or EN).
-2. Then emit **one** fenced JSON patch (and only fields you want to change):
+2. Then emit **one** fenced JSON patch (only fields you want to change):
 
 ```json
 {
@@ -34,14 +37,45 @@ A JSON block named "Live SwarmUI context" is attached. Treat it as ground truth:
   "height": 1280,
   "steps": 8,
   "cfg": 1,
+  "seed": -1,
+  "sigma_shift": 1.15,
+  "sampler": null,
+  "use_init_image": false,
+  "clear_init_image": false,
+  "init_creativity": 0.45,
+  "use_mask_image": false,
+  "clear_mask_image": false,
+  "mask_blur": null,
+  "mask_grow": null,
+  "actions": ["generate"],
+  "search_query": null,
   "notes": "one-line why"
 }
 ```
 
-Rules for the patch:
+### Patch rules
 
 - Omit keys you are not changing.
 - `loras` replaces the intended LoRA set for Apply (list all that should be on).
 - width/height between 128 and 4096; prefer multiples near 1024 for Turbo.
 - Do not invent model or LoRA filenames.
 - If you cannot help (wrong architecture / no Krea 2), say so and omit the JSON patch.
+
+### Init image / inpaint
+
+- **img2img:** set `use_init_image: true` (uses Assistent vision / current Generate image) and `init_creativity` (0 = almost copy, 1 = almost new). Typical edits: **0.25–0.45**; restyle: **0.5–0.7**. Alias `denoise` is accepted.
+- **Inpaint:** needs Init + Mask. Set `use_init_image: true` and `use_mask_image: true` only when the vision pane holds a proper mask (white = edit, black = keep). If the user has not painted a mask, tell them to use Swarm **Edit Image** / paint a mask, or press **As Mask** with a prepared mask — do not invent pixel masks.
+- `clear_init_image` / `clear_mask_image` to leave img2img mode.
+- Prompt Images ≠ Init Image. Prefer Init for structural edits; Prompt Images for style refs (and warn they can dominate).
+
+### Actions (auto-safe)
+
+- `"generate"` — after Apply, start generation (UI auto-generate is on by default).
+- `"use_init"` / `"use_mask"` — same as the boolean flags (optional).
+- `"search_civitai"` — Civitai search; user must **Confirm** downloads.
+- `"interrupt"` — stop generation.
+- Pure Q&A with no prompt/param change: omit the JSON patch entirely (do not burn GPU).
+
+### Auto-apply note
+
+The UI may auto-apply your patch and auto-generate when `actions` contains `generate` or when you change prompt/loras/size/init. Keep patches intentional.

Prompts/compose_scene.md

@@ -8,7 +8,9 @@ Goal: co-create a scene / moodboard direction for **Krea 2**.
 - Propose one strong prompt (not five weak ones).
 - Optionally suggest which available LoRAs fit — only from the live list, with triggers.
 - Mention Prompt Images only if a reference would help, and warn that refs can overpower text.
+- Missing style LoRA → `search_civitai` + `search_query` (Krea-compatible).
 
 ## Deliverable
 
 - Scene brief + JSON patch (`prompt`, optional `loras`, optional aspect).
+- Add `actions: ["generate"]` when ready to try the scene.

Prompts/critique_image.md

@@ -9,8 +9,11 @@ Goal: look at the attached image (vision) and improve the next generation for **
 - If a LoRA trigger was missing or too strong, adjust weight or prompt placement.
 - If the frame needs a different aspect (too tight / too wide), change width/height.
 - Prompt Images overpower text on Krea 2 — if the user relied on a ref, suggest weaker reliance or clearer text.
+- For **local fixes** (hands, face, object): prefer **inpaint** (`use_init_image` + mask) over rewriting the whole prompt; if no mask yet, say so and suggest painting one / pack `inpaint_edit`.
+- For **global restyle**: img2img with moderate `init_creativity` (≈0.4–0.6) can be better than from-scratch.
 
 ## Deliverable
 
 - Short critique in the user's language.
-- JSON patch with improved `prompt` and any `loras` / size tweaks.
+- JSON patch with improved `prompt` and any `loras` / size / init tweaks.
+- Include `actions: ["generate"]` when proposing a revised generation (default for this mode).

Prompts/fix_params.md

@@ -7,10 +7,14 @@ Goal: adjust **generation parameters** for Krea 2 Turbo (or Raw if context says
 - Turbo: prefer steps 4–12 (default 8), CFG ~1, sigma shift ~1.15.
 - Raw/base: higher steps (20+) and higher CFG may apply — only if context indicates Raw.
 - Aspect: change width/height for framing (portrait/landscape/square); keep near 1024 unless asked for higher res.
+- Seed: set `seed` when the user wants reproducibility; `-1` for random.
+- Sampler: only change if the form exposes it and the user asks.
+- **Init creativity** (`init_creativity` / denoise 0–1): only when `has_init_image` or enabling img2img — see pack `inpaint_edit`.
 - Do not change the prompt unless needed to match the new framing.
 - Keep LoRAs unless the user asks to drop them.
 
 ## Deliverable
 
 - Explain the param change.
-- JSON patch focusing on `width`, `height`, `steps`, `cfg` (and `prompt` only if necessary).
+- JSON patch focusing on `width`, `height`, `steps`, `cfg`, `seed`, `sigma_shift`, optional `init_creativity` (and `prompt` only if necessary).
+- Include `actions: ["generate"]` if the user wants to re-roll with the new params.

Prompts/inpaint_edit.md

@@ -0,0 +1,37 @@
+# Mode: inpaint_edit
+
+Goal: guide **img2img** (Init Image) and **inpainting** (Init + Mask) on **Krea 2**.
+
+## When to use which
+
+| Need | Setup |
+| --- | --- |
+| Soft edit / restyle whole frame | Init only + `init_creativity` |
+| Change one region (face, hand, logo) | Init + Mask (white = edit) |
+| Fresh image from text | Clear init/mask; normal t2i |
+
+## Creativity (denoise)
+
+- **0.2–0.35** — small fixes, keep composition
+- **0.4–0.55** — noticeable edit, still related
+- **0.6–0.8** — strong restyle; structure may drift
+- Always set `use_init_image: true` when enabling img2img from the Assistent vision / current image.
+
+## Mask rules
+
+- White = regenerate, black = preserve. Gray = partial.
+- Only set `use_mask_image: true` if context shows a vision image that is meant as a mask, or the user said they prepared one.
+- If `has_mask_image` is false and the user wants regional edit: ask them to paint a mask in Swarm Image Editor (or Assistent **As Mask**), then continue.
+- Optional: `mask_blur` / `mask_grow` for softer edges.
+
+## Prompting
+
+- Describe **what should appear in the edited region**, not the whole scene dump.
+- Keep LoRA triggers if the subject depends on them.
+- Match width/height to the init image when possible.
+
+## Deliverable
+
+- Short plan (img2img vs inpaint) in the user's language.
+- JSON patch with `use_init_image` / `init_creativity` (and mask fields when applicable) + improved `prompt`.
+- `actions: ["generate"]` when ready to run.

Prompts/write_prompt.md

@@ -9,9 +9,11 @@ Goal: craft or improve a **Krea 2** prompt that will generate well on Turbo.
 - Prefer clarity over keyword stuffing. Krea 2 understands sentences.
 - If the user wants a style covered by an available LoRA, enable that LoRA and weave its triggers in.
 - Keep Turbo defaults unless the user asks otherwise (steps 8, cfg 1).
+- If a needed LoRA is missing from `available_loras`, use `actions: ["search_civitai"]` with a clear `search_query` (and prefer Krea base).
 
 ## Deliverable
 
 - Explain briefly what you changed.
 - Emit a JSON patch with at least `prompt`, and `loras` when relevant.
+- Include `actions: ["generate"]` when the user wants to see a new image.
 - Include `width`/`height` only if aspect should change for the scene (e.g. portrait → taller).

README.md

@@ -1,13 +1,13 @@
 # Swarm Assistent
 
-SwarmUI extension for **collaborative Krea 2** prompting via **Ollama**: chat + vision, LoRA/trigger awareness, and applyable prompt/size patches.
+SwarmUI extension for **collaborative Krea 2** prompting via **Ollama**: chat + vision, LoRA/trigger awareness, applyable patches, **img2img / inpaint**, auto Generate, and Civitai search with Confirm.
 
 ## Layout
 
 - **Left:** vision reference (drop / paste / *Send to Assistent* from Generate)
 - **Splitter:** drag to resize panes
 - **Right (wider):** chat
-- **Top-right:** prompt pack + settings (Ollama URL, model, auto-attach)
+- **Top-right:** prompt pack + settings (Ollama URL, model, auto-apply / auto-generate)
 
 ## UX
 
@@ -15,13 +15,19 @@ SwarmUI extension for **collaborative Krea 2** prompting via **Ollama**: chat +
 - Drag images from Generate/History or the OS onto the left pane
 - Paste (`Ctrl+V`) while the Assistent tab is open
 - **Use current** / **Clear** for the vision reference
+- **As Init** / **As Mask** / **Clear Init** — wire the vision image into Swarm `Init Image` / `Mask Image` (img2img & inpaint)
 - Enter sends; Shift+Enter newline
+- **Auto-apply** + **Auto-generate** (default on): patch from the LLM is applied and Generate runs when the patch changes prompt/params or includes `actions: ["generate"]`
+- Pure Q&A without a patch does **not** start Generate
+- **Interrupt** stops Swarm generation / clears busy state
+- **Civitai** search cards require **Confirm download** (uses Swarm `DoModelDownloadWS` + stored `civitai_api` key). Auto-download is off by default.
 
 ## Requirements
 
 - SwarmUI with a **Krea 2** checkpoint selected
-- Ollama on `http://127.0.0.1:11434` (gpu-rent default when `LLM_RUNTIME=ollama`)
-- A chat+vision-capable Ollama model recommended for critique mode
+- Ollama on `http://127.0.0.1:11434` **on the GPU VM** (gpu-rent `LLM_RUNTIME=ollama`). The browser talks to SwarmUI; SwarmUI proxies `/api/tags` and `/api/chat`. URL in settings must stay `127.0.0.1:11434`, not the laptop tunnel port 17811.
+- At least one pulled model (`ollama pull` / `ollama-models.yaml`). Empty `/api/tags` → empty Model dropdown.
+- Optional: Civitai API key in SwarmUI User Settings for search/download.
 
 ## Install
 
@@ -41,13 +47,48 @@ Restart / rebuild SwarmUI after clone.
 
 | Pack | Role |
 | --- | --- |
-| `base_krea2` | Always injected: Krea 2 rules + JSON patch contract |
+| `base_krea2` | Always injected: Krea 2 rules + JSON patch / actions contract |
 | `write_prompt` | Craft / improve prompts |
 | `critique_image` | Vision critique → fixes |
 | `compose_scene` | Scene / moodboard |
-| `fix_params` | Width/height/steps/CFG |
+| `fix_params` | Width/height/steps/CFG/seed/σ-shift |
+| `inpaint_edit` | Init Image img2img + Mask inpaint |
 
-Live context (checkpoint, LoRAs + triggers, current params) is injected every request.
+Live context (checkpoint, server inventory LoRAs + triggers, wildcards, current params) is injected every request.
+
+### Patch actions
+
+```json
+{
+  "prompt": "...",
+  "loras": [{"name": "exact", "weight": 0.8, "triggers": ["..."]}],
+  "width": 1024, "height": 1280, "steps": 8, "cfg": 1,
+  "seed": -1, "sigma_shift": 1.15,
+  "use_init_image": true,
+  "init_creativity": 0.45,
+  "use_mask_image": false,
+  "actions": ["generate"],
+  "search_query": null
+}
+```
+
+- `generate` — auto-generate after apply (when enabled)
+- `use_init` / `use_mask` — set Assistent vision (or current Generate) as Init / Mask
+- `search_civitai` + `search_query` — server searches Civitai, second LLM hop, Confirm cards in UI
+- `interrupt` — stop current generation
+
+Mask convention: **white = edit**, black = keep. Creativity ≈ denoise (0–1).
+
+## API routes
+
+| Route | Role |
+| --- | --- |
+| `AssistentListModels` | Ollama `/api/tags` |
+| `AssistentListInventory` | LoRA / checkpoint / wildcard inventory from Swarm |
+| `AssistentSearchCivitai` | Civitai LoRA search |
+| `AssistentGetPacks` | Prompt pack texts |
+| `AssistentChat` | HTTP chat (+ Civitai hop) |
+| `AssistentChatWS` | Streaming chat WebSocket |
 
 ## License
 

SwarmAssistentExtension.cs

@@ -3,18 +3,22 @@ using System.Collections.Generic;
 using System.IO;
 using System.Linq;
 using System.Net.Http;
+using System.Net.Http.Headers;
+using System.Net.WebSockets;
 using System.Text;
+using System.Text.RegularExpressions;
 using System.Threading.Tasks;
 using FreneticUtilities.FreneticExtensions;
 using Newtonsoft.Json.Linq;
 using SwarmUI.Accounts;
 using SwarmUI.Core;
+using SwarmUI.Text2Image;
 using SwarmUI.Utils;
 using SwarmUI.WebAPI;
 
 namespace Mrleo1nid.SwarmAssistent;
 
-/// <summary>Krea 2 collaborative assistant: Ollama chat + vision + prompt/LoRA/params patches.</summary>
+/// <summary>Krea 2 collaborative assistant: Ollama chat + vision + prompt/LoRA/params patches + Generate/Civitai.</summary>
 public class SwarmAssistentExtension : Extension
 {
     public static PermInfo PermUse = Permissions.Register(new(
@@ -33,17 +37,24 @@ public class SwarmAssistentExtension : Extension
         "critique_image",
         "compose_scene",
         "fix_params",
+        "inpaint_edit",
     ];
 
+    const int MaxCivitaiHops = 2;
+    const int MaxLorasInInventory = 120;
+    const int MaxWildcardsInInventory = 80;
+
+    static readonly Regex JsonFenceRe = new(@"```(?:json)?\s*([\s\S]*?)```", RegexOptions.IgnoreCase | RegexOptions.Compiled);
+
     public override void OnPreInit()
     {
         ScriptFiles.Add("Assets/assistent.js");
         StyleSheetFiles.Add("Assets/assistent.css");
         ExtensionAuthor = "mrleo1nid";
-        Description = "Collaborative Krea 2 assistant via Ollama: chat, vision, prompts, LoRA triggers, size patches.";
+        Description = "Collaborative Krea 2 assistant via Ollama: chat, vision, img2img/inpaint, Generate loop, Civitai Confirm.";
         License = "MIT";
-        Version = "0.2.0";
-        Tags = ["tabs", "ui", "llm", "ollama", "krea"];
+        Version = "0.3.1";
+        Tags = ["tabs", "ui", "llm", "ollama", "krea", "inpaint"];
     }
 
     public override void OnInit()
@@ -51,8 +62,11 @@ public class SwarmAssistentExtension : Extension
         HttpClient ??= new HttpClient { Timeout = TimeSpan.FromMinutes(10) };
         API.RegisterAPICall(AssistentListModels, false, PermUse);
         API.RegisterAPICall(AssistentGetPacks, false, PermUse);
+        API.RegisterAPICall(AssistentListInventory, false, PermUse);
+        API.RegisterAPICall(AssistentSearchCivitai, false, PermUse);
         API.RegisterAPICall(AssistentChat, true, PermUse);
-        Logs.Init("Swarm Assistent extension loaded (Ollama proxy + Krea 2 packs)");
+        API.RegisterAPICall(AssistentChatWS, true, PermUse);
+        Logs.Init("Swarm Assistent extension loaded (Ollama proxy + Krea 2 packs + inventory/Civitai)");
     }
 
     static string Clip(string text, int max)
@@ -128,24 +142,231 @@ public class SwarmAssistentExtension : Extension
         return new JObject { ["success"] = true, ["packs"] = packs, ["order"] = new JArray(PackNames) };
     }
 
-    /// <summary>
-    /// Proxy to Ollama /api/chat (non-stream).
-    /// <paramref name="raw"/> must include messages (JArray) and optional context_json.
-    /// </summary>
-    public async Task<JObject> AssistentChat(Session session, string baseUrl, string model, string pack, bool includeBase, JObject raw)
+    /// <summary>Server-side LoRA / checkpoint / wildcard inventory (not DOM scrape).</summary>
+    public async Task<JObject> AssistentListInventory(Session session)
     {
-        string root = NormalizeBaseUrl(baseUrl ?? raw?["base_url"]?.ToString());
-        string modelName = (model ?? raw?["model"]?.ToString() ?? "").Trim();
-        if (string.IsNullOrWhiteSpace(modelName))
+        await Task.CompletedTask;
+        JArray loras = [];
+        JArray checkpoints = [];
+        JArray wildcards = [];
+
+        if (Program.T2IModelSets.TryGetValue("LoRA", out T2IModelHandler loraHandler))
         {
-            return new JObject { ["error"] = "model is required" };
+            foreach (T2IModel model in loraHandler.Models.Values.OrderBy(m => m.Name).Take(MaxLorasInInventory))
+            {
+                loras.Add(new JObject
+                {
+                    ["name"] = model.Name,
+                    ["title"] = model.Metadata?.Title ?? model.Title ?? model.Name,
+                    ["trigger_phrase"] = model.Metadata?.TriggerPhrase,
+                    ["architecture"] = model.ModelClass?.ID,
+                    ["compat_class"] = model.ModelClass?.CompatClass?.ID,
+                    ["hash"] = model.Metadata?.Hash ?? "",
+                });
             }
-        JArray userMessages = raw?["messages"] as JArray;
-        if (userMessages is null || userMessages.Count == 0)
-        {
-            return new JObject { ["error"] = "messages required" };
         }
 
+        if (Program.T2IModelSets.TryGetValue("Stable-Diffusion", out T2IModelHandler ckptHandler))
+        {
+            foreach (T2IModel model in ckptHandler.Models.Values.OrderBy(m => m.Name).Take(60))
+            {
+                checkpoints.Add(new JObject
+                {
+                    ["name"] = model.Name,
+                    ["title"] = model.Metadata?.Title ?? model.Title ?? model.Name,
+                    ["architecture"] = model.ModelClass?.ID,
+                    ["compat_class"] = model.ModelClass?.CompatClass?.ID,
+                });
+            }
+        }
+
+        try
+        {
+            foreach (string name in WildcardsHelper.ListFiles.OrderBy(n => n).Take(MaxWildcardsInInventory))
+            {
+                wildcards.Add(new JObject { ["name"] = name });
+            }
+        }
+        catch (Exception ex)
+        {
+            Logs.Debug($"AssistentListInventory wildcards: {ex.Message}");
+        }
+
+        bool hasCivitaiKey = !string.IsNullOrWhiteSpace(session.User.GetGenericData("civitai_api", "key"));
+
+        return new JObject
+        {
+            ["success"] = true,
+            ["loras"] = loras,
+            ["checkpoints"] = checkpoints,
+            ["wildcards"] = wildcards,
+            ["has_civitai_key"] = hasCivitaiKey,
+        };
+    }
+
+    /// <summary>Search Civitai for LoRAs (prefers Krea 2 base). Uses Swarm-stored civitai_api key.</summary>
+    public async Task<JObject> AssistentSearchCivitai(Session session, string query, int limit = 8)
+    {
+        string q = (query ?? "").Trim();
+        if (string.IsNullOrWhiteSpace(q))
+        {
+            return new JObject { ["error"] = "query is required" };
+        }
+        limit = Math.Clamp(limit, 1, 20);
+        string apiKey = session.User.GetGenericData("civitai_api", "key") ?? "";
+        HashSet<string> installedNames = CollectInstalledLoraNames();
+        HashSet<string> installedHashes = CollectInstalledLoraHashes();
+
+        string[] hosts = ["civitai.red", "civitai.com"];
+        Exception lastEx = null;
+        foreach (string host in hosts)
+        {
+            try
+            {
+                string url = $"https://{host}/api/v1/models?limit={limit}&types=LORA&query={Uri.EscapeDataString(q)}";
+                using HttpRequestMessage req = new(HttpMethod.Get, url);
+                if (!string.IsNullOrWhiteSpace(apiKey))
+                {
+                    req.Headers.Authorization = new AuthenticationHeaderValue("Bearer", apiKey.Trim());
+                }
+                using HttpResponseMessage resp = await HttpClient.SendAsync(req);
+                string body = await resp.Content.ReadAsStringAsync();
+                if (!resp.IsSuccessStatusCode)
+                {
+                    lastEx = new Exception($"HTTP {(int)resp.StatusCode}: {Clip(body, 200)}");
+                    continue;
+                }
+                JObject parsed = JObject.Parse(body);
+                JArray items = parsed["items"] as JArray ?? [];
+                JArray results = [];
+                foreach (JToken item in items)
+                {
+                    if (item is not JObject mo)
+                    {
+                        continue;
+                    }
+                    JObject card = BuildCivitaiCard(mo, installedNames, installedHashes);
+                    if (card is not null)
+                    {
+                        results.Add(card);
+                    }
+                }
+                // Prefer Krea-compatible first
+                JArray sorted = new(results.OrderByDescending(t => LooksLikeKrea(t["base_model"]?.ToString())).ThenBy(t => t["name"]?.ToString()));
+                return new JObject
+                {
+                    ["success"] = true,
+                    ["query"] = q,
+                    ["host"] = host,
+                    ["results"] = sorted,
+                    ["has_civitai_key"] = !string.IsNullOrWhiteSpace(apiKey),
+                };
+            }
+            catch (Exception ex)
+            {
+                lastEx = ex;
+            }
+        }
+        return new JObject { ["error"] = $"Civitai search failed: {lastEx?.Message ?? "unknown"}" };
+    }
+
+    static bool LooksLikeKrea(string text) => !string.IsNullOrEmpty(text) && Regex.IsMatch(text, @"krea", RegexOptions.IgnoreCase);
+
+    static HashSet<string> CollectInstalledLoraNames()
+    {
+        HashSet<string> names = new(StringComparer.OrdinalIgnoreCase);
+        if (!Program.T2IModelSets.TryGetValue("LoRA", out T2IModelHandler handler))
+        {
+            return names;
+        }
+        foreach (T2IModel m in handler.Models.Values)
+        {
+            names.Add(m.Name);
+            string leaf = m.Name.Replace('\\', '/').AfterLast('/');
+            if (!string.IsNullOrEmpty(leaf))
+            {
+                names.Add(leaf);
+                names.Add(Path.GetFileNameWithoutExtension(leaf));
+            }
+        }
+        return names;
+    }
+
+    static HashSet<string> CollectInstalledLoraHashes()
+    {
+        HashSet<string> hashes = new(StringComparer.OrdinalIgnoreCase);
+        if (!Program.T2IModelSets.TryGetValue("LoRA", out T2IModelHandler handler))
+        {
+            return hashes;
+        }
+        foreach (T2IModel m in handler.Models.Values)
+        {
+            string h = m.Metadata?.Hash;
+            if (!string.IsNullOrWhiteSpace(h))
+            {
+                hashes.Add(h.Trim().ToLowerInvariant());
+            }
+        }
+        return hashes;
+    }
+
+    static JObject BuildCivitaiCard(JObject model, HashSet<string> installedNames, HashSet<string> installedHashes)
+    {
+        string name = model["name"]?.ToString() ?? "";
+        JArray versions = model["modelVersions"] as JArray;
+        JObject ver = versions?.FirstOrDefault() as JObject;
+        if (ver is null)
+        {
+            return null;
+        }
+        string baseModel = ver["baseModel"]?.ToString() ?? "";
+        JArray trained = ver["trainedWords"] as JArray ?? [];
+        List<string> triggers = trained.Select(t => t?.ToString()).Where(s => !string.IsNullOrWhiteSpace(s)).Take(8).ToList();
+        JObject file = null;
+        foreach (JToken f in ver["files"] as JArray ?? [])
+        {
+            if (f is JObject fo && (fo["primary"]?.Value<bool>() == true || (fo["name"]?.ToString() ?? "").EndsWith(".safetensors", StringComparison.OrdinalIgnoreCase)))
+            {
+                file = fo;
+                break;
+            }
+        }
+        file ??= (ver["files"] as JArray)?.FirstOrDefault() as JObject;
+        string downloadUrl = file?["downloadUrl"]?.ToString() ?? ver["downloadUrl"]?.ToString() ?? "";
+        string fileName = file?["name"]?.ToString() ?? "";
+        string sha = file?["hashes"]?["SHA256"]?.ToString() ?? file?["hashes"]?["AutoV2"]?.ToString() ?? "";
+        string saveName = string.IsNullOrWhiteSpace(fileName)
+            ? Regex.Replace(name, @"[^\w\-.]+", "_").Trim('_')
+            : Path.GetFileNameWithoutExtension(fileName);
+
+        bool already = false;
+        if (!string.IsNullOrWhiteSpace(sha) && installedHashes.Contains(sha.Trim().ToLowerInvariant()))
+        {
+            already = true;
+        }
+        else if (installedNames.Contains(saveName) || installedNames.Contains(name) || installedNames.Contains(fileName))
+        {
+            already = true;
+        }
+
+        return new JObject
+        {
+            ["id"] = model["id"],
+            ["version_id"] = ver["id"],
+            ["name"] = name,
+            ["base_model"] = baseModel,
+            ["krea_likely"] = LooksLikeKrea(baseModel),
+            ["triggers"] = new JArray(triggers),
+            ["download_url"] = downloadUrl,
+            ["file_name"] = saveName,
+            ["sha256"] = sha,
+            ["already_installed"] = already,
+            ["n_sfw"] = model["nsfw"]?.Value<bool>() ?? false,
+        };
+    }
+
+    List<JObject> BuildOllamaMessages(string packName, bool includeBase, string contextJson, JArray userMessages, string extraSystem = null)
+    {
         List<JObject> ollamaMessages = [];
         StringBuilder system = new();
         if (includeBase)
@@ -156,7 +377,6 @@ public class SwarmAssistentExtension : Extension
                 system.AppendLine(basePack);
             }
         }
-        string packName = (pack ?? raw?["pack"]?.ToString() ?? "write_prompt").Trim();
         if (!string.IsNullOrWhiteSpace(packName) && packName != "base_krea2")
         {
             string situational = ReadPackFile(packName);
@@ -167,7 +387,6 @@ public class SwarmAssistentExtension : Extension
                 system.AppendLine(situational);
             }
         }
-        string contextJson = raw?["context_json"]?.ToString();
         if (!string.IsNullOrWhiteSpace(contextJson))
         {
             system.AppendLine();
@@ -176,6 +395,11 @@ public class SwarmAssistentExtension : Extension
             system.AppendLine(contextJson);
             system.AppendLine("```");
         }
+        if (!string.IsNullOrWhiteSpace(extraSystem))
+        {
+            system.AppendLine();
+            system.AppendLine(extraSystem);
+        }
         if (system.Length > 0)
         {
             ollamaMessages.Add(new JObject
@@ -184,7 +408,7 @@ public class SwarmAssistentExtension : Extension
                 ["content"] = system.ToString(),
             });
         }
-        foreach (JToken msg in userMessages)
+        foreach (JToken msg in userMessages ?? [])
         {
             if (msg is not JObject mo)
             {
@@ -201,24 +425,264 @@ public class SwarmAssistentExtension : Extension
             }
             ollamaMessages.Add(copy);
         }
+        return ollamaMessages;
+    }
 
+    static JObject TryParsePatch(string reply)
+    {
+        if (string.IsNullOrWhiteSpace(reply))
+        {
+            return null;
+        }
+        foreach (Match match in JsonFenceRe.Matches(reply))
+        {
+            string raw = match.Groups[1].Value.Trim();
+            try
+            {
+                JObject obj = JObject.Parse(raw);
+                if (obj is not null && (obj["prompt"] != null || obj["loras"] != null || obj["width"] != null
+                    || obj["height"] != null || obj["steps"] != null || obj["cfg"] != null
+                    || obj["seed"] != null || obj["sigma_shift"] != null || obj["sampler"] != null
+                    || obj["actions"] != null || obj["search_query"] != null || obj["civitai_query"] != null
+                    || obj["use_init_image"] != null || obj["clear_init_image"] != null
+                    || obj["init_creativity"] != null || obj["denoise"] != null
+                    || obj["use_mask_image"] != null || obj["clear_mask_image"] != null
+                    || obj["mask_blur"] != null || obj["mask_grow"] != null))
+                {
+                    return obj;
+                }
+            }
+            catch
+            {
+                // not json
+            }
+        }
+        return null;
+    }
+
+    static string ExtractSearchQuery(JObject patch)
+    {
+        if (patch is null)
+        {
+            return null;
+        }
+        string q = (patch["search_query"] ?? patch["civitai_query"])?.ToString()?.Trim();
+        if (!string.IsNullOrWhiteSpace(q))
+        {
+            return q;
+        }
+        if (patch["actions"] is JArray acts)
+        {
+            foreach (JToken a in acts)
+            {
+                if (string.Equals(a?.ToString(), "search_civitai", StringComparison.OrdinalIgnoreCase))
+                {
+                    return q; // may still be null — caller checks
+                }
+            }
+        }
+        return null;
+    }
+
+    static bool WantsCivitaiSearch(JObject patch)
+    {
+        if (patch is null)
+        {
+            return false;
+        }
+        if (!string.IsNullOrWhiteSpace(ExtractSearchQuery(patch)))
+        {
+            return true;
+        }
+        if (patch["actions"] is JArray acts)
+        {
+            foreach (JToken a in acts)
+            {
+                if (string.Equals(a?.ToString(), "search_civitai", StringComparison.OrdinalIgnoreCase))
+                {
+                    return true;
+                }
+            }
+        }
+        return false;
+    }
+
+    async Task<(string reply, JObject raw, JArray civitaiResults)> RunChatWithHops(
+        Session session,
+        string root,
+        string modelName,
+        string packName,
+        bool includeBase,
+        string contextJson,
+        JArray userMessages,
+        Func<string, Task> onDelta = null,
+        Func<int, Task> onHopStart = null)
+    {
+        List<JObject> messages = BuildOllamaMessages(packName, includeBase, contextJson, userMessages);
+        JArray civitaiResults = [];
+        string reply = "";
+        JObject lastRaw = null;
+        for (int hop = 0; hop < MaxCivitaiHops; hop++)
+        {
+            if (onHopStart is not null)
+            {
+                await onHopStart(hop);
+            }
+            (reply, lastRaw) = await CallOllamaChat(root, modelName, messages, stream: onDelta is not null, onDelta);
+            JObject patch = TryParsePatch(reply);
+            if (hop + 1 >= MaxCivitaiHops || !WantsCivitaiSearch(patch))
+            {
+                break;
+            }
+            string query = ExtractSearchQuery(patch);
+            if (string.IsNullOrWhiteSpace(query))
+            {
+                query = userMessages.LastOrDefault(m => m["role"]?.ToString() == "user")?["content"]?.ToString() ?? "";
+            }
+            if (string.IsNullOrWhiteSpace(query))
+            {
+                break;
+            }
+            JObject search = await AssistentSearchCivitai(session, query, 8);
+            if (search["error"] is not null)
+            {
+                messages.Add(new JObject { ["role"] = "assistant", ["content"] = reply });
+                messages.Add(new JObject
+                {
+                    ["role"] = "user",
+                    ["content"] = $"Civitai search failed: {search["error"]}. Continue without download — use only available_loras from context.",
+                });
+                continue;
+            }
+            civitaiResults = search["results"] as JArray ?? [];
+            messages.Add(new JObject { ["role"] = "assistant", ["content"] = reply });
+            messages.Add(new JObject
+            {
+                ["role"] = "user",
+                ["content"] =
+                    "Civitai search results (JSON). Prefer `krea_likely: true`. Do NOT download yourself — the UI shows Confirm cards. " +
+                    "Pick useful LoRAs from results or available_loras, emit a normal patch (prompt/loras). " +
+                    "Omit search_civitai from actions unless you need a different query.\n```json\n" +
+                    civitaiResults.ToString(Newtonsoft.Json.Formatting.None) + "\n```",
+            });
+        }
+        return (reply, lastRaw, civitaiResults);
+    }
+
+    async Task<(string reply, JObject raw)> CallOllamaChat(
+        string root,
+        string modelName,
+        List<JObject> ollamaMessages,
+        bool stream,
+        Func<string, Task> onDelta)
+    {
         JObject payload = new()
         {
             ["model"] = modelName,
-            ["stream"] = false,
+            ["stream"] = stream,
             ["messages"] = new JArray(ollamaMessages),
         };
-        try
-        {
         using StringContent content = new(payload.ToString(Newtonsoft.Json.Formatting.None), Encoding.UTF8, "application/json");
-            using HttpResponseMessage resp = await HttpClient.PostAsync($"{root}/api/chat", content);
-            string body = await resp.Content.ReadAsStringAsync();
+        using HttpRequestMessage req = new(HttpMethod.Post, $"{root}/api/chat") { Content = content };
+        using HttpResponseMessage resp = await HttpClient.SendAsync(req, stream
+            ? HttpCompletionOption.ResponseHeadersRead
+            : HttpCompletionOption.ResponseContentRead);
         if (!resp.IsSuccessStatusCode)
         {
-                return new JObject { ["error"] = $"Ollama /api/chat HTTP {(int)resp.StatusCode}: {Clip(body, 800)}" };
+            string errBody = await resp.Content.ReadAsStringAsync();
+            throw new Exception($"Ollama /api/chat HTTP {(int)resp.StatusCode}: {Clip(errBody, 800)}");
         }
+        if (!stream)
+        {
+            string body = await resp.Content.ReadAsStringAsync();
             JObject parsed = JObject.Parse(body);
             string reply = parsed["message"]?["content"]?.ToString() ?? parsed["response"]?.ToString() ?? "";
+            return (reply, parsed);
+        }
+        StringBuilder full = new();
+        await using Stream streamBody = await resp.Content.ReadAsStreamAsync();
+        using StreamReader reader = new(streamBody, Encoding.UTF8);
+        JObject last = null;
+        while (true)
+        {
+            string line = await reader.ReadLineAsync();
+            if (line is null)
+            {
+                break;
+            }
+            if (string.IsNullOrWhiteSpace(line))
+            {
+                continue;
+            }
+            JObject chunk = JObject.Parse(line);
+            last = chunk;
+            string delta = chunk["message"]?["content"]?.ToString() ?? "";
+            if (!string.IsNullOrEmpty(delta))
+            {
+                full.Append(delta);
+                if (onDelta is not null)
+                {
+                    await onDelta(delta);
+                }
+            }
+            if (chunk["done"]?.Value<bool>() == true)
+            {
+                break;
+            }
+        }
+        return (full.ToString(), last ?? new JObject());
+    }
+
+    /// <summary>
+    /// SwarmUI passes the whole request as the JObject param (not only a nested key).
+    /// Support both flat fields and legacy nested <c>raw</c>.
+    /// </summary>
+    static void ExtractChatPayload(JObject raw, ref string baseUrl, ref string model, ref string pack, ref bool includeBase, out JArray userMessages, out string contextJson)
+    {
+        JObject whole = raw ?? [];
+        JObject nested = whole["raw"] as JObject;
+        if (string.IsNullOrWhiteSpace(baseUrl))
+        {
+            baseUrl = whole["base_url"]?.ToString()
+                ?? whole["baseUrl"]?.ToString()
+                ?? nested?["base_url"]?.ToString()
+                ?? nested?["baseUrl"]?.ToString();
+        }
+        if (string.IsNullOrWhiteSpace(model))
+        {
+            model = whole["model"]?.ToString() ?? nested?["model"]?.ToString();
+        }
+        if (string.IsNullOrWhiteSpace(pack))
+        {
+            pack = whole["pack"]?.ToString() ?? nested?["pack"]?.ToString();
+        }
+        if (whole["includeBase"] is not null)
+        {
+            includeBase = whole.Value<bool?>("includeBase") ?? includeBase;
+        }
+        userMessages = (whole["messages"] as JArray) ?? (nested?["messages"] as JArray);
+        contextJson = whole["context_json"]?.ToString() ?? nested?["context_json"]?.ToString();
+    }
+
+    /// <summary>Proxy to Ollama /api/chat (non-stream), with optional Civitai search hop.</summary>
+    public async Task<JObject> AssistentChat(Session session, string baseUrl, string model, string pack, bool includeBase, JObject raw)
+    {
+        ExtractChatPayload(raw, ref baseUrl, ref model, ref pack, ref includeBase, out JArray userMessages, out string contextJson);
+        string root = NormalizeBaseUrl(baseUrl);
+        string modelName = (model ?? "").Trim();
+        if (string.IsNullOrWhiteSpace(modelName))
+        {
+            return new JObject { ["error"] = "model is required" };
+        }
+        if (userMessages is null || userMessages.Count == 0)
+        {
+            return new JObject { ["error"] = "messages required" };
+        }
+        string packName = (pack ?? "write_prompt").Trim();
+        try
+        {
+            (string reply, JObject parsed, JArray civitai) = await RunChatWithHops(
+                session, root, modelName, packName, includeBase, contextJson, userMessages);
             return new JObject
             {
                 ["success"] = true,
@@ -226,6 +690,7 @@ public class SwarmAssistentExtension : Extension
                 ["model"] = modelName,
                 ["pack"] = packName,
                 ["raw"] = parsed,
+                ["civitai_results"] = civitai,
             };
         }
         catch (Exception ex)
@@ -233,4 +698,62 @@ public class SwarmAssistentExtension : Extension
             return new JObject { ["error"] = $"Ollama chat failed: {ex.Message}" };
         }
     }
+
+    /// <summary>WebSocket streaming chat (Ollama stream:true) + Civitai hops.</summary>
+    public async Task<JObject> AssistentChatWS(Session session, WebSocket ws, string baseUrl, string model, string pack, bool includeBase, JObject raw)
+    {
+        ExtractChatPayload(raw, ref baseUrl, ref model, ref pack, ref includeBase, out JArray userMessages, out string contextJson);
+        string root = NormalizeBaseUrl(baseUrl);
+        string modelName = (model ?? "").Trim();
+        if (string.IsNullOrWhiteSpace(modelName))
+        {
+            await ws.SendJson(new JObject { ["error"] = "model is required" }, API.WebsocketTimeout);
+            return null;
+        }
+        if (userMessages is null || userMessages.Count == 0)
+        {
+            await ws.SendJson(new JObject { ["error"] = "messages required" }, API.WebsocketTimeout);
+            return null;
+        }
+        string packName = (pack ?? "write_prompt").Trim();
+        try
+        {
+            async Task OnDelta(string delta)
+            {
+                if (ws.State == WebSocketState.Open)
+                {
+                    await ws.SendJson(new JObject { ["delta"] = delta }, API.WebsocketTimeout);
+                }
+            }
+            async Task OnHopStart(int hop)
+            {
+                if (ws.State == WebSocketState.Open && hop > 0)
+                {
+                    await ws.SendJson(new JObject
+                    {
+                        ["clear_stream"] = true,
+                        ["hop"] = hop + 1,
+                        ["notice"] = "Civitai search done — refining…",
+                    }, API.WebsocketTimeout);
+                }
+            }
+            (string reply, JObject parsed, JArray civitai) = await RunChatWithHops(
+                session, root, modelName, packName, includeBase, contextJson, userMessages, OnDelta, OnHopStart);
+            await ws.SendJson(new JObject
+            {
+                ["success"] = true,
+                ["done"] = true,
+                ["reply"] = reply,
+                ["model"] = modelName,
+                ["pack"] = packName,
+                ["raw"] = parsed,
+                ["civitai_results"] = civitai,
+            }, API.WebsocketTimeout);
+        }
+        catch (Exception ex)
+        {
+            await ws.SendJson(new JObject { ["error"] = $"Ollama chat failed: {ex.Message}" }, API.WebsocketTimeout);
+        }
+        return null;
+    }
 }

Tabs/Text2Image/Assistent.html

@@ -15,6 +15,9 @@
             </div>
             <div class="sa-image-actions">
                 <button type="button" class="basic-button" id="sa_btn_use_current" title="Grab the current Generate preview">Use current</button>
+                <button type="button" class="basic-button" id="sa_btn_as_init" title="Set vision image as Swarm Init Image (img2img)">As Init</button>
+                <button type="button" class="basic-button" id="sa_btn_as_mask" title="Set vision image as Mask (inpaint — white=edit)">As Mask</button>
+                <button type="button" class="basic-button" id="sa_btn_clear_init" title="Clear Init Image + Mask on Generate tab">Clear Init</button>
                 <button type="button" class="basic-button" id="sa_btn_clear_image" title="Clear vision image">Clear</button>
                 <label class="sa-check"><input type="checkbox" id="sa_attach_vision" checked /> Attach next</label>
             </div>
@@ -29,23 +32,30 @@
                         <option value="critique_image">Critique image</option>
                         <option value="compose_scene">Compose scene</option>
                         <option value="fix_params">Fix params</option>
+                        <option value="inpaint_edit">Inpaint / img2img</option>
+                    </select>
+                    <select id="sa_model" class="sa-select sa-model-select" title="Ollama model">
+                        <option value="">Loading models…</option>
                     </select>
                     <button type="button" class="basic-button sa-icon-btn" id="sa_btn_settings" title="Settings" aria-label="Settings">⚙</button>
                 </div>
             </header>
             <div class="sa-settings" id="sa_settings" hidden>
                 <label>Ollama URL <input type="text" id="sa_base_url" value="http://127.0.0.1:11434" /></label>
-                <label>Model
-                    <select id="sa_model" class="sa-select"></select>
-                </label>
                 <button type="button" class="basic-button" id="sa_btn_refresh_models">Refresh models</button>
+                <button type="button" class="basic-button" id="sa_btn_refresh_inventory">Refresh inventory</button>
                 <label class="sa-check"><input type="checkbox" id="sa_auto_vision" /> Auto-attach current image</label>
+                <label class="sa-check"><input type="checkbox" id="sa_auto_apply" checked /> Auto-apply patch</label>
+                <label class="sa-check"><input type="checkbox" id="sa_auto_generate" checked /> Auto-generate after patch</label>
+                <label class="sa-check"><input type="checkbox" id="sa_auto_critique" /> Auto-critique after generate</label>
+                <label class="sa-check sa-danger" title="Dangerous — downloads without Confirm"><input type="checkbox" id="sa_auto_download" /> Auto-download Civitai (off)</label>
             </div>
             <div class="sa-messages" id="sa_messages"></div>
             <div class="sa-composer">
-                <textarea id="sa_input" rows="3" placeholder="Ask for a prompt, critique the image, change aspect… (Enter = send, Shift+Enter = newline)"></textarea>
+                <textarea id="sa_input" rows="3" placeholder="Ask for a prompt, img2img, inpaint, critique… (Enter = send, Shift+Enter = newline)"></textarea>
                 <div class="sa-composer-actions">
                     <button type="button" class="basic-button sa-primary" id="sa_btn_send">Send</button>
+                    <button type="button" class="basic-button" id="sa_btn_interrupt" title="Interrupt generation / chat" hidden>Interrupt</button>
                     <button type="button" class="basic-button" id="sa_btn_clear">Clear chat</button>
                     <span class="sa-status" id="sa_status"></span>
                 </div>