Add exact KV memory with persona overlays and param priority.

Canonical defaults live in exact.json (UI fills empties without an LLM call); chat session overrides beat file defaults until clear/persona change. Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-21 23:04:34 +03:00
parent fe4d8d3a3a
commit 1b9cc1ad78
15 changed files with 418 additions and 121 deletions
@@ -1,6 +1,6 @@
 /**
 * Swarm Assistent — Krea 2 collaborative chat (Ollama via Swarm API).
- * v0.7.0: Config presets, persona folders, vector memory, chat|memory model roles.
+ * v0.7.0: Config presets, persona folders, exact KV + vector memory, chat|memory model roles.
 */
 (function () {
    const LS_BASE = 'swarm_assistent_base_url';
@@ -99,6 +99,9 @@
        history: [],
        packsLoaded: false,
        config: null,
+        exact: null,
+        sessionExact: {},
+        lastUserParamIntent: false,
        enabledSkills: [],
        kreaProfiles: { turbo: { steps: 8, cfg: 1, sigma_shift: 1.15 }, raw: { steps: 28, cfg: 4.5 } },
        preferredEmbed: null,
@@ -424,6 +427,158 @@
        el.dispatchEvent(new Event('change', { bubbles: true }));
    }

+    function isEmptyParamField(raw, { treatZeroEmpty = false } = {}) {
+        if (raw == null) {
+            return true;
+        }
+        const s = String(raw).trim();
+        if (s === '') {
+            return true;
+        }
+        if (treatZeroEmpty && (s === '0' || Number(s) === 0)) {
+            return true;
+        }
+        return false;
+    }
+
+    function userTextMentionsParams(text) {
+        return /\b(steps?|cfg|seed|sigma|размер|aspect|ширин|высот|resolution|batch|турбо|turbo|raw)\b/i.test(String(text || ''));
+    }
+
+    function applyAspectTableFrom(obj) {
+        if (!obj || typeof obj !== 'object') {
+            return false;
+        }
+        const next = {};
+        for (const [k, v] of Object.entries(obj)) {
+            if (Array.isArray(v) && v.length >= 2) {
+                next[k] = [Number(v[0]), Number(v[1])];
+            }
+        }
+        if (!Object.keys(next).length) {
+            return false;
+        }
+        ASPECT_TABLE = next;
+        return true;
+    }
+
+    function resolveExactBundle() {
+        const exact = state.exact || state.config?.exact || {};
+        const profiles = exact.profiles || state.kreaProfiles || {};
+        return { exact, profiles };
+    }
+
+    function detectKreaProfileName() {
+        try {
+            const model = resolveCurrentCheckpoint();
+            const blob = `${model?.name || ''} ${model?.title || ''}`.toLowerCase();
+            const hasRaw = /\braw\b/.test(blob);
+            const hasTurbo = /\bturbo\b/.test(blob);
+            return hasRaw && !hasTurbo ? 'raw' : 'turbo';
+        } catch (e) {
+            return (state.exact?.generation?.profile) || 'turbo';
+        }
+    }
+
+    function mergedGenerationDefaults(profileName) {
+        const { exact, profiles } = resolveExactBundle();
+        const gen = exact.generation && typeof exact.generation === 'object' ? { ...exact.generation } : {};
+        const profile = profileName || gen.profile || detectKreaProfileName();
+        const fromProfile = profiles[profile] && typeof profiles[profile] === 'object' ? { ...profiles[profile] } : {};
+        const session = state.sessionExact && typeof state.sessionExact === 'object' ? { ...state.sessionExact } : {};
+        // Profile (turbo/raw) overrides generation defaults; session overrides both.
+        return { ...gen, ...fromProfile, profile, ...session };
+    }
+
+    function exactDefaultFor(key, profileName) {
+        const { exact, profiles } = resolveExactBundle();
+        const profile = profileName || exact.generation?.profile || detectKreaProfileName();
+        const fromProfile = profiles[profile]?.[key];
+        if (fromProfile != null) {
+            return fromProfile;
+        }
+        return exact.generation?.[key];
+    }
+
+    function rememberSessionExact(partial) {
+        if (!partial || typeof partial !== 'object') {
+            return;
+        }
+        const keys = ['steps', 'cfg', 'sigma_shift', 'aspect', 'width', 'height', 'images', 'batch', 'seed', 'sampler', 'scheduler'];
+        for (const k of keys) {
+            if (partial[k] != null) {
+                state.sessionExact[k] = partial[k];
+            }
+        }
+        if (partial.images == null && partial.batch != null) {
+            state.sessionExact.images = partial.batch;
+        }
+    }
+
+    function fillEmptyParamsFromExact() {
+        const defaults = mergedGenerationDefaults();
+        if (isEmptyParamField(val('input_steps'), { treatZeroEmpty: true }) && defaults.steps != null) {
+            setVal('input_steps', String(defaults.steps));
+        }
+        const cfgRaw = val('input_cfgscale') || val('input_cfg');
+        if (isEmptyParamField(cfgRaw, { treatZeroEmpty: true }) && defaults.cfg != null) {
+            if (document.getElementById('input_cfgscale')) {
+                setVal('input_cfgscale', String(defaults.cfg));
+            } else if (document.getElementById('input_cfg')) {
+                setVal('input_cfg', String(defaults.cfg));
+            }
+        }
+        if (isEmptyParamField(val('input_sigmashift')) && defaults.sigma_shift != null) {
+            setVal('input_sigmashift', String(defaults.sigma_shift));
+        }
+        const wEmpty = isEmptyParamField(val('input_width'), { treatZeroEmpty: true });
+        const hEmpty = isEmptyParamField(val('input_height'), { treatZeroEmpty: true });
+        if ((wEmpty || hEmpty) && defaults.aspect) {
+            const size = sizeFromAspect(defaults.aspect);
+            if (size) {
+                if (wEmpty) {
+                    setVal('input_width', String(size[0]));
+                }
+                if (hEmpty) {
+                    setVal('input_height', String(size[1]));
+                }
+            }
+        } else {
+            if (wEmpty && defaults.width != null) {
+                setVal('input_width', String(defaults.width));
+            }
+            if (hEmpty && defaults.height != null) {
+                setVal('input_height', String(defaults.height));
+            }
+        }
+        const batchId = document.getElementById('input_images') ? 'input_images' : (document.getElementById('input_batchsize') ? 'input_batchsize' : null);
+        if (batchId && isEmptyParamField(val(batchId), { treatZeroEmpty: true })) {
+            const batch = defaults.images != null ? defaults.images : defaults.batch;
+            if (batch != null) {
+                setVal(batchId, String(batch));
+            }
+        }
+    }
+
+    function shouldSkipSessionRollback(key, patchValue) {
+        if (state.lastUserParamIntent) {
+            return false;
+        }
+        if (state.sessionExact[key] == null) {
+            return false;
+        }
+        const sessionVal = state.sessionExact[key];
+        if (String(sessionVal) === String(patchValue)) {
+            return false;
+        }
+        const exactVal = exactDefaultFor(key);
+        if (exactVal == null) {
+            return false;
+        }
+        // Model trying to restore file exact while session override differs — keep session.
+        return String(patchValue) === String(exactVal);
+    }
+
    function openAssistentTab() {
        const tab = document.getElementById(TAB_BUTTON_ID);
        if (tab) {
@@ -1143,6 +1298,8 @@

    function onPersonaChanged() {
        const id = $('sa_persona')?.value || 'neutral';
+        state.sessionExact = {};
+        state.lastUserParamIntent = false;
        saveSettings();
        loadConfig(id, (data) => {
            const title = data?.personas?.find((p) => p.id === id)?.title
@@ -1159,6 +1316,7 @@
                    $('sa_pack').value = packId;
                }
            }
+            fillEmptyParamsFromExact();
        });
    }

@@ -1480,15 +1638,32 @@
            const blob = `${model.name || ''} ${model.title || ''}`.toLowerCase();
            const hasRaw = /\braw\b/.test(blob);
            const hasTurbo = /\bturbo\b/.test(blob);
-            ctx.krea_profile = hasRaw && !hasTurbo ? 'raw' : 'turbo';
-            ctx.recommended_params = ctx.krea_profile === 'raw'
-                ? { steps: 28, cfg: 4.5 }
-                : { steps: 8, cfg: 1, sigma_shift: 1.15 };
+            const profile = hasRaw && !hasTurbo ? 'raw' : 'turbo';
+            ctx.krea_profile = profile;
+            const defaults = mergedGenerationDefaults(profile);
+            ctx.recommended_params = {
+                steps: defaults.steps ?? 8,
+                cfg: defaults.cfg ?? 1,
+                sigma_shift: defaults.sigma_shift ?? 1.15,
+            };
+            if (defaults.aspect) {
+                ctx.recommended_params.aspect = defaults.aspect;
+            }
        } catch (e) {
            ctx.krea_profile = 'turbo';
-            ctx.recommended_params = { steps: 8, cfg: 1, sigma_shift: 1.15 };
+            const defaults = mergedGenerationDefaults('turbo');
+            ctx.recommended_params = {
+                steps: defaults.steps ?? 8,
+                cfg: defaults.cfg ?? 1,
+                sigma_shift: defaults.sigma_shift ?? 1.15,
+            };
        }

+        ctx.exact = state.exact || state.config?.exact || null;
+        ctx.session_exact = state.sessionExact && Object.keys(state.sessionExact).length
+            ? { ...state.sessionExact }
+            : {};
+
        return ctx;
    }

@@ -1901,7 +2076,7 @@
        if (/\b(inpaint|замажь|закрась|руки|лицо|маск|mask|img2img|init\s*image)\b/i.test(t)) {
            return 'inpaint_edit';
        }
-        if (/\b(steps?|cfg|seed|sigma|размер|aspect|ширин|высот|resolution|batch)\b/i.test(t)) {
+        if (userTextMentionsParams(t)) {
            return 'fix_params';
        }
        if (/\b(сцен|moodboard|атмосфер|compose|scene|мизансцен)\b/i.test(t)) {
@@ -2010,27 +2185,68 @@
        }

        if (doParams) {
+            const defaults = mergedGenerationDefaults();
+            const capture = state.lastUserParamIntent;
            const aspectSize = sizeFromAspect(patch.aspect);
+            if (patch.aspect != null && !shouldSkipSessionRollback('aspect', patch.aspect)) {
                if (aspectSize) {
                    setVal('input_width', String(aspectSize[0]));
                    setVal('input_height', String(aspectSize[1]));
+                }
+                if (capture) {
+                    rememberSessionExact({ aspect: patch.aspect });
+                }
+            } else if (patch.aspect == null && isEmptyParamField(val('input_width'), { treatZeroEmpty: true })
+                && isEmptyParamField(val('input_height'), { treatZeroEmpty: true }) && defaults.aspect) {
+                const fill = sizeFromAspect(defaults.aspect);
+                if (fill) {
+                    setVal('input_width', String(fill[0]));
+                    setVal('input_height', String(fill[1]));
+                }
            } else {
-                if (patch.width != null) {
+                if (patch.width != null && !shouldSkipSessionRollback('width', patch.width)) {
                    setVal('input_width', String(patch.width));
+                    if (capture) {
+                        rememberSessionExact({ width: patch.width });
                    }
-                if (patch.height != null) {
+                } else if (patch.width == null && isEmptyParamField(val('input_width'), { treatZeroEmpty: true }) && defaults.width != null) {
+                    setVal('input_width', String(defaults.width));
+                }
+                if (patch.height != null && !shouldSkipSessionRollback('height', patch.height)) {
                    setVal('input_height', String(patch.height));
+                    if (capture) {
+                        rememberSessionExact({ height: patch.height });
+                    }
+                } else if (patch.height == null && isEmptyParamField(val('input_height'), { treatZeroEmpty: true }) && defaults.height != null) {
+                    setVal('input_height', String(defaults.height));
                }
            }
-            if (patch.steps != null) {
+            if (patch.steps != null && !shouldSkipSessionRollback('steps', patch.steps)) {
                setVal('input_steps', String(patch.steps));
+                if (capture) {
+                    rememberSessionExact({ steps: patch.steps });
                }
-            if (patch.cfg != null) {
+            } else if (patch.steps == null && isEmptyParamField(val('input_steps'), { treatZeroEmpty: true }) && defaults.steps != null) {
+                setVal('input_steps', String(defaults.steps));
+            }
+            if (patch.cfg != null && !shouldSkipSessionRollback('cfg', patch.cfg)) {
                if (document.getElementById('input_cfgscale')) {
                    setVal('input_cfgscale', String(patch.cfg));
                } else {
                    setVal('input_cfg', String(patch.cfg));
                }
+                if (capture) {
+                    rememberSessionExact({ cfg: patch.cfg });
+                }
+            } else if (patch.cfg == null) {
+                const cfgRaw = val('input_cfgscale') || val('input_cfg');
+                if (isEmptyParamField(cfgRaw, { treatZeroEmpty: true }) && defaults.cfg != null) {
+                    if (document.getElementById('input_cfgscale')) {
+                        setVal('input_cfgscale', String(defaults.cfg));
+                    } else if (document.getElementById('input_cfg')) {
+                        setVal('input_cfg', String(defaults.cfg));
+                    }
+                }
            }
            if (patch.vary === true) {
                setVal('input_seed', '-1');
@@ -2039,27 +2255,50 @@
                if (cur && String(cur) !== '-1') {
                    setVal('input_seed', cur);
                }
-            } else if (patch.seed != null) {
+            } else if (patch.seed != null && !shouldSkipSessionRollback('seed', patch.seed)) {
                setVal('input_seed', String(patch.seed));
+                if (capture) {
+                    rememberSessionExact({ seed: patch.seed });
                }
-            if (patch.sigma_shift != null) {
+            }
+            if (patch.sigma_shift != null && !shouldSkipSessionRollback('sigma_shift', patch.sigma_shift)) {
                setVal('input_sigmashift', String(patch.sigma_shift));
+                if (capture) {
+                    rememberSessionExact({ sigma_shift: patch.sigma_shift });
+                }
+            } else if (patch.sigma_shift == null && isEmptyParamField(val('input_sigmashift')) && defaults.sigma_shift != null) {
+                setVal('input_sigmashift', String(defaults.sigma_shift));
            }
            if (patch.sampler != null) {
                if (document.getElementById('input_sampler')) {
                    setVal('input_sampler', String(patch.sampler));
                }
+                if (capture) {
+                    rememberSessionExact({ sampler: patch.sampler });
+                }
            }
            if (patch.scheduler != null && document.getElementById('input_scheduler')) {
                setVal('input_scheduler', String(patch.scheduler));
+                if (capture) {
+                    rememberSessionExact({ scheduler: patch.scheduler });
+                }
            }
            const batch = patch.images != null ? patch.images : patch.batch;
-            if (batch != null) {
+            if (batch != null && !shouldSkipSessionRollback('images', batch)) {
                if (document.getElementById('input_images')) {
                    setVal('input_images', String(batch));
                } else if (document.getElementById('input_batchsize')) {
                    setVal('input_batchsize', String(batch));
                }
+                if (capture) {
+                    rememberSessionExact({ images: batch });
+                }
+            } else if (batch == null) {
+                const batchId = document.getElementById('input_images') ? 'input_images' : (document.getElementById('input_batchsize') ? 'input_batchsize' : null);
+                const defBatch = defaults.images != null ? defaults.images : defaults.batch;
+                if (batchId && isEmptyParamField(val(batchId), { treatZeroEmpty: true }) && defBatch != null) {
+                    setVal(batchId, String(defBatch));
+                }
            }
        }

@@ -2835,19 +3074,16 @@
            return;
        }
        state.config = data;
-        if (data.model?.aspect_table && typeof data.model.aspect_table === 'object') {
-            const next = {};
-            for (const [k, v] of Object.entries(data.model.aspect_table)) {
-                if (Array.isArray(v) && v.length >= 2) {
-                    next[k] = [Number(v[0]), Number(v[1])];
+        if (data.exact && typeof data.exact === 'object') {
+            state.exact = data.exact;
        }
+        const aspectSource = data.exact?.aspect_table || data.model?.aspect_table;
+        if (aspectSource && typeof aspectSource === 'object') {
+            applyAspectTableFrom(aspectSource);
        }
-            if (Object.keys(next).length) {
-                ASPECT_TABLE = next;
-            }
-        }
-        if (data.model?.profiles) {
-            state.kreaProfiles = data.model.profiles;
+        const profileSource = data.exact?.profiles || data.model?.profiles;
+        if (profileSource && typeof profileSource === 'object') {
+            state.kreaProfiles = profileSource;
        }
        if (data.ui?.pack_aliases) {
            PACK_ALIASES = { ...PACK_ALIASES, ...data.ui.pack_aliases };
@@ -2879,6 +3115,9 @@
        if (data.assistant?.embed_model && !state.preferredEmbed) {
            state.preferredEmbed = data.assistant.embed_model;
        }
+        if (applyDefaults || data.exact) {
+            fillEmptyParamsFromExact();
+        }
    }

    function renderPersonaOptions(personas, selected) {
@@ -3958,7 +4197,10 @@
        if (!Array.isArray(withActions.actions) && patchHasGenTrigger(withActions)) {
            withActions.actions = ['generate'];
        }
+        const prevIntent = state.lastUserParamIntent;
+        state.lastUserParamIntent = true;
        await applyPatch(withActions, 'all');
+        state.lastUserParamIntent = prevIntent;
        setStatus(note || 'Applied');
        if ($('sa_auto_generate')?.checked) {
            await runGenerateFromPatch(withActions);
@@ -4209,6 +4451,9 @@
        if (!text) {
            return;
        }
+        if (!opts.fromAutoCritique && !opts.fromDownload && !opts.fromVisionHop) {
+            state.lastUserParamIntent = userTextMentionsParams(text);
+        }

        if (!opts.fromAutoCritique && !opts.fromDownload && !opts.fromVisionHop && !opts.skipSlash) {
            if (rawInput.startsWith('/')) {
@@ -4745,6 +4990,8 @@
            state.visionHopUsed = false;
            state.packUserTouched = false;
            state.pendingPersonaNote = null;
+            state.sessionExact = {};
+            state.lastUserParamIntent = false;
            clearPersistedHistory();
            const box = $('sa_messages');
            if (box) {
@@ -4786,11 +5033,11 @@
            } else if (vary) {
                await applyQuickPatch({ vary: true, seed: -1, actions: ['generate'] }, 'Vary');
            } else if (profile === 'turbo') {
-                const p = state.kreaProfiles?.turbo || { steps: 8, cfg: 1, sigma_shift: 1.15 };
+                const p = state.kreaProfiles?.turbo || mergedGenerationDefaults('turbo');
                await applyQuickPatch({ steps: p.steps ?? 8, cfg: p.cfg ?? 1, sigma_shift: p.sigma_shift ?? 1.15, actions: ['generate'] }, 'Turbo');
            } else if (profile === 'raw') {
-                const p = state.kreaProfiles?.raw || { steps: 28, cfg: 4.5 };
-                await applyQuickPatch({ steps: p.steps ?? 28, cfg: p.cfg ?? 4.5, actions: ['generate'] }, 'RAW');
+                const p = state.kreaProfiles?.raw || mergedGenerationDefaults('raw');
+                await applyQuickPatch({ steps: p.steps ?? 28, cfg: p.cfg ?? 4.5, sigma_shift: p.sigma_shift, actions: ['generate'] }, 'RAW');
            }
            renderLoraChips();
        });
@@ -319,6 +319,9 @@ public sealed class AssistentConfig

    public JObject LoadUi(string personaId) => MergeJsonLayers("ui.json", LayerRoots(personaId));

+    /// <summary>Exact (KV) memory: bundled base → persona overlays → disk overlays. Persona keys overwrite base.</summary>
+    public JObject LoadExact(string personaId) => MergeJsonLayers("exact.json", LayerRoots(personaId));
+
    public JObject LoadModelProfile(string personaId)
    {
        JObject assistant = LoadAssistant(personaId);
@@ -712,6 +715,7 @@ public sealed class AssistentConfig
        JObject assistant = LoadAssistant(id);
        JObject ui = LoadUi(id);
        JObject model = LoadModelProfile(id);
+        JObject exact = LoadExact(id);
        var packs = ListPacks(id);
        var skills = ListSkills(id);
        var personas = ListPersonaCatalog();
@@ -724,6 +728,7 @@ public sealed class AssistentConfig
            ["assistant"] = assistant,
            ["ui"] = ui,
            ["model"] = model,
+            ["exact"] = exact,
            ["packs"] = new JArray(packs.Select(p => new JObject
            {
                ["id"] = p.id,
@@ -10,7 +10,7 @@
  "default_persona": "neutral",
  "embed_model": "nomic-embed-text",
  "memory_top_k": 10,
-  "seed_version": 1,
+  "seed_version": 2,
  "gate": {
    "architecture": "krea2",
    "keywords": ["krea"]
@@ -2,13 +2,28 @@

 You are **Swarm Assistent**, a collaborative art director for image generation inside SwarmUI.

+## Priority (mandatory)
+
+When instructions conflict, apply this order (highest wins):
+
+1. **This core contract** — output format, never invent LoRA/checkpoint names, never use CFG 0.
+2. **Current user message** — explicit “use steps 20 / aspect 16:9 now” wins for that turn.
+3. **Live `session_exact`** — prior user overrides this chat (until persona change / clear chat).
+4. **Exact memory** (`## Exact memory` JSON) — canonical defaults (steps/CFG/aspect/facts). Persona overlays are already merged into it.
+5. **Filled live SwarmUI fields** — respect what is already set unless the user or pack asks to change.
+6. **`memory_hits` (vector RAG)** — notes, pitfalls, LoRA blurbs. Never override exact numbers or the user’s param request.
+7. Guesses — last resort only.
+
+Exact = encyclopedia of defaults. RAG = soft notes. Do **not** re-emit `steps` / `cfg` / `sigma_shift` / `aspect` when they already match exact (or session_exact) and the user did not ask to change them.
+
 ## Live context

 A JSON block named "Live SwarmUI context" is attached. Treat it as ground truth — refreshed every chat turn:

 - Use only LoRAs listed in `available_loras` / `enabled_loras` (exact `name`), or Civitai search candidates.
 - Prefer listed `trigger_phrase` / `triggers` — **never invent** trigger words.
- `memory_hits` are retrieved facts (model knowledge, LoRA notes, pitfalls). Trust them over guesses.
+- `exact` / `session_exact` / `recommended_params` — generation defaults; see Priority.
+- `memory_hits` are retrieved notes (LoRA tips, pitfalls). Trust them over guesses, but **not** over exact or the user.
 - `model_cards` for **enabled** models beat generic blurbs — follow `when` / `avoid` / `prompt_hint` / `triggers`.
 - `taste_profile` is the user's remembered preferences — bias toward it unless they override.
 - Prefer `krea_likely` / Krea architecture entries; ignore FLUX/SDXL LoRAs.
@@ -67,6 +82,7 @@ A JSON block named "Live SwarmUI context" is attached. Treat it as ground truth
 ### Patch rules

 - Omit keys you are not changing.
+- Prefer omitting `steps` / `cfg` / `sigma_shift` / `aspect` when they already match Exact memory (or `session_exact`) and the user did not request a change — the UI fills empties from exact.
 - `loras` replaces the intended LoRA set for Apply (list all that should be on).
 - Prefer `aspect` over raw width/height when framing changes.
 - `vary: true` — new random seed. `lock_seed: true` — reuse current seed.
@@ -0,0 +1,38 @@
+{
+  "generation": {
+    "profile": "turbo",
+    "steps": 8,
+    "cfg": 1,
+    "sigma_shift": 1.15,
+    "images": 1
+  },
+  "profiles": {
+    "turbo": {
+      "steps": 8,
+      "cfg": 1,
+      "sigma_shift": 1.15
+    },
+    "raw": {
+      "steps": 28,
+      "cfg": 4.5,
+      "sigma_shift": 1.15
+    }
+  },
+  "aspect_table": {
+    "1:1": [1024, 1024],
+    "4:3": [1184, 896],
+    "3:2": [1248, 832],
+    "16:9": [1376, 768],
+    "2.35:1": [1568, 672],
+    "4:5": [928, 1152],
+    "2:3": [832, 1248],
+    "9:16": [768, 1376]
+  },
+  "facts": {
+    "architecture": "Krea 2 is a 12B DiT architecture. Not FLUX, not SDXL, not FLUX.1-Krea. Text encoder: Qwen3-VL 4B. VAE: Qwen Image VAE. Use only Krea2-trained LoRAs — never suggest FLUX/SDXL LoRAs.",
+    "negatives": "Negative prompts are nearly useless with Qwen3-VL. Prefer positives (sharp focus, empty street) over no blur / no people. Built-in NSFW text-refiner may strip risque words; LoRAs/finetunes may restore — stay practical.",
+    "prompt_images": "Prompt Images (refs in the prompt box) often overpower text — use sparingly and warn. Init Image = structure (img2img). Mask = local fix. They are not interchangeable. Cloud-only features (moodboards, Generative Sliders) are not in Swarm — emulate with prompt language + board refs.",
+    "turbo": "Krea 2 Turbo defaults live in exact.generation / exact.profiles.turbo. Never use CFG 0 — broken output. Side ~1024 (128–4096 OK).",
+    "raw": "Krea 2 RAW/Base: prefer exact.profiles.raw when checkpoint name/title looks like RAW (not Turbo). If a turbo LoRA exists, weight ~0.6 for photoreal (1.0 ≈ full turbo). Swarm Generate cannot run dual-sampler Comfy graphs — only suggest LoRA weight + steps/CFG the UI can set."
+  }
+}
@@ -1,50 +1,8 @@
 [
  {
    "kind": "aspect",
-    "key": "1:1",
+    "key": "table",
    "tags": ["aspect", "1k"],
-    "text": "Aspect 1:1 maps to 1024×1024 on the official Krea 1K table. Prefer patch field aspect over raw width/height."
-  },
-  {
-    "kind": "aspect",
-    "key": "4:3",
-    "tags": ["aspect", "1k"],
-    "text": "Aspect 4:3 maps to 1184×896."
-  },
-  {
-    "kind": "aspect",
-    "key": "3:2",
-    "tags": ["aspect", "1k"],
-    "text": "Aspect 3:2 maps to 1248×832."
-  },
-  {
-    "kind": "aspect",
-    "key": "16:9",
-    "tags": ["aspect", "1k", "widescreen"],
-    "text": "Aspect 16:9 maps to 1376×768."
-  },
-  {
-    "kind": "aspect",
-    "key": "2.35:1",
-    "tags": ["aspect", "1k", "cinematic"],
-    "text": "Aspect 2.35:1 (cinematic ultrawide) maps to 1568×672."
-  },
-  {
-    "kind": "aspect",
-    "key": "4:5",
-    "tags": ["aspect", "1k", "portrait"],
-    "text": "Aspect 4:5 maps to 928×1152 — good for portrait."
-  },
-  {
-    "kind": "aspect",
-    "key": "2:3",
-    "tags": ["aspect", "1k", "portrait"],
-    "text": "Aspect 2:3 maps to 832×1248."
-  },
-  {
-    "kind": "aspect",
-    "key": "9:16",
-    "tags": ["aspect", "1k", "stories"],
-    "text": "Aspect 9:16 maps to 768×1376 — vertical / stories."
+    "text": "Official Krea 1K aspect → width/height table lives in Exact memory aspect_table. Prefer patch field aspect over raw width/height. UI fills empties from Exact."
  }
 ]
@@ -3,30 +3,30 @@
    "kind": "model",
    "key": "krea2_architecture",
    "tags": ["krea", "architecture"],
-    "text": "Krea 2 is a 12B DiT architecture. Not FLUX, not SDXL, not FLUX.1-Krea. Text encoder: Qwen3-VL 4B. VAE: Qwen Image VAE. Use only Krea2-trained LoRAs — never suggest FLUX/SDXL LoRAs."
+    "text": "Krea 2 architecture facts live in Exact memory (exact.facts.architecture). Prefer Exact KV over this note. Never suggest FLUX/SDXL LoRAs."
  },
  {
    "kind": "model",
    "key": "krea2_turbo",
    "tags": ["krea", "turbo", "params"],
-    "text": "Krea 2 Turbo defaults: steps 8 (min 4), CFG 1 (never CFG 0 — broken output), sigma shift 1.15, side ~1024 (128–4096 OK)."
+    "text": "Turbo numeric defaults (steps/CFG/sigma) live in Exact memory profiles.turbo / generation — do not invent numbers; never use CFG 0."
  },
  {
    "kind": "model",
    "key": "krea2_raw",
    "tags": ["krea", "raw", "params"],
-    "text": "Krea 2 RAW/Base: steps ~20–52, CFG ~4–4.5. If checkpoint name/title looks like RAW (not Turbo), prefer RAW settings. If a turbo LoRA exists, weight ~0.6 for photoreal (1.0 ≈ full turbo). Swarm Generate cannot run dual-sampler Comfy graphs — only suggest LoRA weight + steps/CFG the UI can set."
+    "text": "RAW numeric defaults live in Exact memory profiles.raw. If checkpoint looks like RAW (not Turbo), use that profile. Turbo LoRA weight ~0.6 for photoreal when needed."
  },
  {
    "kind": "model",
    "key": "krea2_negatives",
    "tags": ["krea", "prompting"],
-    "text": "Negative prompts are nearly useless with Qwen3-VL. Prefer positives (sharp focus, empty street) over no blur / no people. Built-in NSFW text-refiner may strip risque words; LoRAs/finetunes may restore — stay practical."
+    "text": "See Exact memory facts.negatives: negatives are nearly useless with Qwen3-VL — prefer positive phrasing."
  },
  {
    "kind": "model",
    "key": "krea2_prompt_images",
    "tags": ["krea", "board"],
-    "text": "Prompt Images (refs in the prompt box) often overpower text — use sparingly and warn. Init Image = structure (img2img). Mask = local fix. They are not interchangeable. Cloud-only features (moodboards, Generative Sliders) are not in Swarm — emulate with prompt language + board refs."
+    "text": "See Exact memory facts.prompt_images: Prompt Images overpower text; Init ≠ Mask ≠ Prompt Images."
  }
 ]
@@ -1,26 +1,4 @@
 {
  "id": "krea2",
-  "gate_keywords": ["krea"],
-  "profiles": {
-    "turbo": {
-      "steps": 8,
-      "cfg": 1,
-      "sigma_shift": 1.15
-    },
-    "raw": {
-      "steps": 28,
-      "cfg": 4.5,
-      "sigma_shift": 1.15
-    }
-  },
-  "aspect_table": {
-    "1:1": [1024, 1024],
-    "4:3": [1184, 896],
-    "3:2": [1248, 832],
-    "16:9": [1376, 768],
-    "2.35:1": [1568, 672],
-    "4:5": [928, 1152],
-    "2:3": [832, 1248],
-    "9:16": [768, 1376]
-  }
+  "gate_keywords": ["krea"]
 }
@@ -4,16 +4,16 @@ Goal: adjust **generation parameters** for Krea 2 Turbo (or RAW if context says

 ## Guidelines

- **Turbo:** steps 4–12 (default **8**), CFG **1** (never 0), sigma shift ~**1.15**.
- **RAW/base:** steps 20–52, CFG ~4–4.5 — only if checkpoint/context indicates Raw. If a turbo-distill LoRA is available, weight **0.6** is the usual photoreal compromise (UI LoRA only — no dual-sampler).
- Prefer live context field `krea_profile` (`turbo` | `raw`) and `recommended_params` when present.
- **Aspect:** prefer patch field `aspect` (`1:1`, `4:5`, `2:3`, `16:9`, `9:16`, `4:3`, `3:2`, `2.35:1`) — UI maps to official 1K sizes. Else set width/height near 1024.
+- Prefer **Exact memory** (`profiles.turbo` / `profiles.raw`), live `recommended_params`, and `session_exact` over invented numbers. Never CFG 0.
+- Prefer live context field `krea_profile` (`turbo` | `raw`) when present.
+- **Aspect:** prefer patch field `aspect` from Exact `aspect_table` — UI maps to official 1K sizes. Else set width/height near 1024.
 - **Batch:** `images` or `batch` (1–4 typical).
 - **Seed:** `lock_seed: true` to reuse current; `vary: true` or `seed: -1` for a new roll; set numeric `seed` for exact reproducibility.
 - **Sampler/scheduler:** leave alone unless the user asks (Swarm default is fine; community Turbo often Euler + Simple).
 - **Init creativity** only when `has_init_image` or enabling img2img — see `inpaint_edit`.
 - Do not change the prompt unless needed for the new framing.
 - Keep LoRAs unless asked to drop them.
+- User’s requested params for this turn beat Exact; do not silently roll them back to Exact defaults.

 ## Deliverable

@@ -15,7 +15,7 @@ Structure as flowing prose (not tag soup):

 - Put **LoRA trigger phrases** near the subject they affect.
 - Short user ideas → expand. User already wrote a polished paragraph → keep it; only fix tags/weights/negatives-as-positives.
- Keep Turbo defaults unless asked (steps 8, cfg 1). Prefer `aspect` for framing.
+- Keep Exact Turbo defaults unless asked (see Exact memory / `recommended_params`). Prefer `aspect` for framing. Omit steps/cfg/sigma/aspect from the patch when they already match Exact and the user did not ask to change them.
 - Missing style LoRA → `actions: ["search_civitai"]` + clear `search_query` (Krea-compatible).
 - Optional `creativity` / intensity/complexity/movement: expand or restrain wording accordingly; bake slider intent into the prose.

@@ -1,6 +1,6 @@
 {
  "id": "memory",
-  "title": "Vector memory",
+  "title": "Exact + vector memory",
  "default": true,
  "prompt_file": "memory.md"
 }
@@ -1,8 +1,15 @@
 # Skill: memory

-You have a persistent vector memory (`memory_hits` in live context).
+You have two memory layers:

-## When to write
+1. **Exact memory** (`## Exact memory` + live `exact` / `session_exact`) — canonical KV defaults (generation params, aspect table, architecture facts). Always prefer Exact over RAG for numbers and defaults.
+2. **Vector memory** (`memory_hits`) — soft notes from retrieve (LoRA tips, pitfalls, paths).
+
+## Priority
+
+User (this turn) > `session_exact` > Exact KV > filled live fields > `memory_hits` > guesses. Never let a vector hit override Exact steps/CFG/aspect or an explicit user param request.
+
+## When to write (vector only)

 - Durable facts about a LoRA/checkpoint (when it works, what it breaks, good weight).
 - Bad paths / pitfalls you discovered this session.
@@ -10,6 +17,7 @@ You have a persistent vector memory (`memory_hits` in live context).

 ## When not to write

+- Do not dump Exact defaults into vector memory — they already live in `exact.json`.
 - Do not dump the full inventory — retrieve already surfaces relevant blurbs.
 - Do not store the user's taste profile (that is `taste_profile` / taste.json).
 - Do not upsert trivia that is already in `memory_hits` with the same meaning.
@@ -0,0 +1,5 @@
+{
+  "generation": {
+    "aspect": "2.35:1"
+  }
+}
@@ -2,7 +2,7 @@

 SwarmUI extension for **collaborative Krea 2** prompting via **Ollama**: chat + board (Generate | Refs tabs), LoRA chips, **persona presets** (`Config/personas/`), **vector memory**, model cards with Civitai fetch, img2img/inpaint, slash commands, auto Generate.

-**Version 0.7.0** — Config/_base + persona folders, skills, memory-seed → SQLite, Ollama `use: chat|memory`, slim inventory via retrieve.
+**Version 0.7.0** — Config/_base + persona folders, skills, **exact KV memory** + vector memory-seed → SQLite, Ollama `use: chat|memory`, slim inventory via retrieve.

 ## Layout

@@ -15,20 +15,29 @@ SwarmUI extension for **collaborative Krea 2** prompting via **Ollama**: chat +

 ```
 Config/
-  _base/           # defaults (assistant, ui, models/krea2, core, packs, skills, memory-seed, identity)
-  personas/<id>/   # sparse preset: persona/voice/likes/dislikes/rules + optional overrides
+  _base/           # defaults (assistant, ui, models/krea2, exact.json, core, packs, skills, memory-seed, identity)
+  personas/<id>/   # sparse preset: persona/voice/likes/dislikes/rules + optional exact.json / overrides
 ```

 Disk overlay (wins over bundled): `/mnt/swarm_data/Assistent/` — same layout, plus `settings.json`, `taste.json`, `personas.json` (legacy prompt overlay), `ollama-roles.json`, `memory/assistent.sqlite`.

 Copy `personas/cinema/` → `noir/`, edit only differing JSON files.

+## Exact memory (KV)
+
+- `Config/_base/exact.json` — canonical generation defaults, profiles (turbo/raw), aspect table, short facts
+- Persona / disk overlays merge via DeepMerge (matching keys overwrite)
+- Always injected into the system prompt; UI fills **empty** SwarmUI fields from Exact (no LLM call)
+- Chat-session overrides (`session_exact`) last until persona change or clear chat — not written to disk
+- Priority: core contract → current user → session_exact → exact (+ persona) → live fields → vector `memory_hits`
+
 ## Vector memory

 - SQLite + Ollama `/api/embed` (default `nomic-embed-text`, pick in ⚙)
- First chat seeds `Config/_base/memory-seed/` (Krea facts, aspect, pitfalls)
+- First chat seeds `Config/_base/memory-seed/` (pointers + pitfalls; numbers live in Exact)
 - Agents upsert via patch `memory_upsert` / `memory_forget`
 - Cards ingest on save; retrieve → `memory_hits` in live context (inventory slimmed)
+- Soft notes only — Exact and the user beat RAG for params

 ## UX

@@ -76,7 +85,7 @@ Restart / rebuild SwarmUI after clone. gpu-rent: `seed-extensions` + restart.

 **Packs** (one active): `write_prompt`, `critique_image`, `compose_scene`, `fix_params`, `inpaint_edit`, `describe_ref`, `catalog_card`.

-**Skills** (checkboxes): `prompting`, `creativity_sliders`, `memory` — procedures, not model encyclopedia (facts live in memory-seed).
+**Skills** (checkboxes): `prompting`, `creativity_sliders`, `memory` — procedures; encyclopedia numbers live in Exact, soft notes in memory-seed / RAG.

 **Personas:** `neutral`, `lewd`, `aggressive`, `cinema`, `terse` under `Config/personas/`.

@@ -1347,6 +1347,16 @@ public class SwarmAssistentExtension : Extension
            }
        }

+        JObject exact = Config.LoadExact(pid);
+        if (exact is not null && exact.Count > 0)
+        {
+            system.AppendLine();
+            system.AppendLine("## Exact memory (canonical KV — always trust over RAG guesses)");
+            system.AppendLine("```json");
+            system.AppendLine(exact.ToString(Newtonsoft.Json.Formatting.None));
+            system.AppendLine("```");
+        }
+
        foreach (string skillId in skillIds ?? Config.ResolveEnabledSkills(pid, null))
        {
            string skillText = Config.LoadSkillPrompt(pid, skillId);
@@ -1564,7 +1574,7 @@ public class SwarmAssistentExtension : Extension
            Logs.Debug($"Assistent memory retrieve: {ex.Message}");
        }

-        string enrichedContext = InjectMemoryHits(contextJson, hits);
+        string enrichedContext = InjectMemoryHits(contextJson, hits, Config.LoadExact(pid));
        List<JObject> messages = BuildOllamaMessages(packName, includeBase, enrichedContext, userMessages, personaId: pid, skillIds: skills);
        JArray civitaiResults = [];
        string reply = "";
@@ -1659,7 +1669,7 @@ public class SwarmAssistentExtension : Extension
        return string.IsNullOrWhiteSpace(q) ? "krea2 prompting" : q;
    }

-    static string InjectMemoryHits(string contextJson, JArray hits)
+    static string InjectMemoryHits(string contextJson, JArray hits, JObject exact = null)
    {
        JObject ctx;
        try
@@ -1671,6 +1681,29 @@ public class SwarmAssistentExtension : Extension
            ctx = new JObject { ["_raw_context"] = contextJson };
        }
        ctx["memory_hits"] = hits ?? new JArray();
+        if (exact is not null && exact.Count > 0 && ctx["exact"] is null)
+        {
+            ctx["exact"] = exact;
+        }
+        if (ctx["session_exact"] is null)
+        {
+            ctx["session_exact"] = new JObject();
+        }
+        if (ctx["recommended_params"] is null && exact?["generation"] is JObject gen)
+        {
+            JObject rec = new();
+            foreach (string key in new[] { "steps", "cfg", "sigma_shift", "aspect", "images" })
+            {
+                if (gen[key] is not null)
+                {
+                    rec[key] = gen[key].DeepClone();
+                }
+            }
+            if (rec.Count > 0)
+            {
+                ctx["recommended_params"] = rec;
+            }
+        }
        // Slim inventory for LLM: keep enabled + current, drop full dump if present
        if (ctx["available_loras"] is JArray allLoras && allLoras.Count > 24)
        {