Leonid PershinandCursor
5f33130381
Client always merges Exact turbo|raw steps/cfg/sigma before Generate so sparse LLM omissions and leftover SD 20/7 cannot stick; Ollama park/warm skip no-op /api/ps round-trips when the chat model is already unloaded or resident. Co-authored-by: Cursor <cursoragent@cursor.com>
20 lines
1.2 KiB
Markdown
20 lines
1.2 KiB
Markdown
# Mode: compose_scene
|
|
|
|
Goal: co-create a scene / moodboard direction for **Krea 2** (local Swarm).
|
|
|
|
## Approach
|
|
|
|
- Clarify subject, setting, time of day, camera distance, style.
|
|
- Propose **one** strong prompt (not five weak ones).
|
|
- **Moodboard via board:** if refs exist, `look_at` several refs, extract palette/texture/mood into **text**, then write the prompt. Prefer text distillation over dumping refs as Prompt Images.
|
|
- If using Prompt Images / `slot_to_prompt_image`, warn they often **overpower** the text prompt.
|
|
- Suggest available LoRAs only from the live list, with triggers.
|
|
- Missing style LoRA → ask the user or use `"ask": ["inventory"]` for names already on disk (Krea-compatible).
|
|
- Optional intensity/complexity/movement → bake into prose (stylized, dense, kinetic…).
|
|
|
|
## Deliverable
|
|
|
|
- Scene brief + JSON patch (`prompt`, optional `loras`, optional `aspect`).
|
|
- `actions: ["generate"]` / `"generate": true` when ready to try the scene.
|
|
- Omitting `steps` / `cfg` / `sigma_shift` is safe — the client fills Exact turbo|raw for the live checkpoint. Prefer including them when live session disagrees with `krea_profile` (keeps the session honest).
|