Runpod Pixel Art

If ai/.env is missing, stop and ask the user to create it from ai/.env.example.

Default Workflow

1. Read ai/README.md and the requested workflow JSON in ai/workflows/.
2. If the user does not already have a workflow, start from ai/workflows/example-pixel-sprite-workflow.json.
3. Confirm any model-specific placeholder values are replaced, especially ckpt_name.
4. Run ./ai/scripts/runpod-start.sh.
5. Run ./ai/scripts/runpod-wait.sh.
6. Run ./ai/scripts/runpod-status.sh if you need the current URL, cost, or mapped ports.
7. Submit the workflow with ./ai/scripts/comfy-submit.sh <workflow-json>.
8. Fetch outputs with ./ai/scripts/comfy-fetch.sh <prompt-id> <output-dir>.
9. If appropriate, stop the pod with ./ai/scripts/runpod-stop.sh.

Bash Commands

Use these commands from the workspace root:

./ai/scripts/runpod-create.sh
./ai/scripts/runpod-start.sh
./ai/scripts/runpod-wait.sh
./ai/scripts/runpod-status.sh
prompt_id="$(./ai/scripts/comfy-submit.sh ai/workflows/example-pixel-sprite-workflow.json)"
./ai/scripts/comfy-fetch.sh "$prompt_id" ai/output/example
./ai/scripts/runpod-stop.sh

Single-command path:

./ai/scripts/generate.sh ai/workflows/example-pixel-sprite-workflow.json ai/output/example

Iteration Rules

• Keep workflows small and debuggable at first.
• Prefer one checkpoint and one prompt target at a time.
• Save outputs under descriptive folders in ai/output/.
• Change one axis at a time when iterating: prompt, seed, size, CFG, or steps.
• For pixel art, prefer cleaner prompts over higher CFG or more steps.

Asset Guidance

Good first targets:

• one player concept sprite
• one crawler enemy
• one drone enemy
• one pickup icon set
• one tile mood test

Typical prompt constraints:

• pixel art
• sprite asset
• transparent background
• centered composition
• readable silhouette
• limited palette
• retro console game look

Typical negative constraints:

• photorealistic
• painterly
• blurry
• detailed background
• watermark
• text
• multiple characters

Failure Handling

If pod creation or start fails:

• run ./ai/scripts/runpod-status.sh
• inspect ai/.env
• verify the GPU type string is valid for the current RunPod API

If ComfyUI never becomes ready:

• verify the image or template actually launches ComfyUI on 0.0.0.0:8188
• verify COMFYUI_PORT and COMFYUI_HEALTH_PATH

If submission succeeds but outputs are empty:

• inspect the workflow JSON
• verify ckpt_name
• lower complexity before adding LoRAs or custom nodes