Generate scene bg
Generate ADV / Visual Novel style scene background illustrations. Use when creating narrative game backdrops, dialogue scene backgrounds, VN/ADV location art, or scene illustrations in anime, semi-realistic, painted, or cel-shaded styles. Supports optional time-of-day variants (day/night/sunset/rain) with visual reference handoff for consistency. NOT for character or sprite assets (use generate-tachie).From its SKILL.md
npx -y skills add tznthou/tachie-forge --skill generate-scene-bgAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
4.6 KB, 986 tokens by cl100k_base, as published. Nobody here has run it
Generate Scene BG
Generate scene background illustrations for ADV / Visual Novel style games (Persona, Ace Attorney, etc.). Output is a finished single-image backdrop, not a layered game map.
Parameters
Infer from the user request. Only prompt is required.
art_style:anime_illustration|semi_realistic|painted_scene|cel_shaded|project-nativeprompt: the user's scene descriptionaspect_ratio:16:9(default) |4:3| custom WxHtime_variant: optional —day|sunset|night|rain|overcast| custom
When unspecified:
- Default
art_styletoanime_illustration. - Default
aspect_ratioto16:9(1920x1080). - Omit
time_variantunless the user asks for a specific time or weather.
Workflow
Step 1 — Scene Plan
From the user's description, infer:
- What location this is (classroom, courtroom, café, rooftop, etc.)
- Art style and aspect ratio
- Whether a time variant is requested
- Output file naming:
assets/bg/<scene-name>.png
State the plan briefly and confirm before generating.
Step 2 — Generate Background
- Write the image prompt manually. The prompt must describe:
- The scene composition, perspective, and depth
- Art style cues matching the chosen
art_style - Lighting and atmosphere matching
time_variantif set - "No characters, no text, no UI elements" — backgrounds must be empty of actors
- Call
image_genwith the prompt. If the agent has no nativeimage_gentool (e.g. Claude Code), invoke it via Bash instead:scripts/call-codex-imagegen.sh -o assets/bg/<scene-name>.png "<prompt>"— this bridges to Codex CLI'simage_gen. - Save the prompt as
assets/bg/<scene-name>.prompt.txt. - Validate: correct dimensions/aspect ratio, no baked-in characters or text.
Generating Variants (same scene, different time/angle)
When the user asks for a variant of an existing background (e.g., "same classroom but at night"):
- Make the original background visible in conversation context. If it is a local file, call
view_imageimmediately before writing the variant prompt. - In the variant prompt, explicitly say: use the visible image above as the visual reference — preserve the room layout, furniture positions, perspective, and composition. Change only the lighting/atmosphere/weather to match the new variant.
- Generate the variant and save as
assets/bg/<scene-name>-<variant>.pngwith its own.prompt.txt. - Save the base prompt alongside the variant prompt so the lineage is traceable.
This is best-effort consistency — image generation does not guarantee identical composition across variants. The visual reference handoff improves consistency but does not make it pixel-perfect.
Prompt Guidelines
Art Style Cues
| Style | Prompt cues |
|---|---|
anime_illustration | anime-style background illustration, detailed anime scenery, soft lighting, clean lines |
semi_realistic | semi-realistic digital painting, detailed environment art, cinematic lighting |
painted_scene | hand-painted background, watercolor-inspired, painterly atmosphere |
cel_shaded | cel-shaded background, flat color with clean outlines, stylized shading |
What to Include
- Perspective and camera angle (eye-level, slightly elevated, wide shot, etc.)
- Key architectural/environmental features
- Lighting direction and quality
- Atmosphere and mood through visual description (not a separate mood parameter — describe it in the prompt)
- Time of day through lighting and sky cues when
time_variantis set
What to Exclude
- Characters, people, silhouettes, or figures
- Text, signs with readable text, UI elements
- Game HUD or overlay elements
File Structure
assets/bg/
classroom.png
classroom.prompt.txt
classroom-night.png
classroom-night.prompt.txt
courtroom.png
courtroom.prompt.txt
Validation
- Output image matches requested
aspect_ratio - No characters or text baked into the background
- Prompt file saved alongside every generated image
- Variant filenames follow
<scene-name>-<variant>.pngconvention
Boundaries
- Character sprites, standing illustrations →
generate-tachie - This skill produces scene illustrations only — no runtime metadata, no engine wiring
What ships with it: 1 file
886 B alongside SKILL.md
agents/
- openai.yaml886 B