Game Art ComfyUI Generator

Problem

Game art generation often gets stuck in repetitive ComfyUI operations: character portraits, scene concept art, prop icons, and VFX frames each need different model settings, resolutions, and prompt constraints. When engineers manually load API workflows, replace variables, and submit /prompt tasks, they can easily introduce inconsistent parameters, exceed VRAM limits, or lose track of output paths.

How It Works

This skill centralizes local ComfyUI configuration: set COMFYUI_BASE_URL, optional COMFYUI_API_KEY, and COMFYUI_OUTPUT_DIR, then maintain character, scene, prop, and vfx workflow templates. Each template is a ComfyUI API JSON with default CheckpointLoaderSimple, KSampler, CLIPTextEncode, VAEDecode, and SaveImage nodes, making it easier to tune sampler, steps, and size by asset type.

At runtime it performs three steps:
- Preflight checks: verify that the ComfyUI service is reachable, an asset template exists, and batch_count stays within max_batch_size, splitting batches or adjusting resolution when needed.
- Parameter replacement: write prompt, negative_prompt, width, height, seed, and batch_size into {{variable}} placeholders in the selected template.
- Task submission: send a POST request to COMFYUI_BASE_URL/prompt, include Authorization: Bearer if a key is configured, and submit the replaced workflow JSON with a fixed client_id.

Boundaries

It fits local ComfyUI environments where teams need batch image generation organized by game asset type. The defaults assume an SDXL example model and RTX 3090 VRAM thresholds, so local models, node plugins, and output directories must be available. It is not a general post-processing tool; transparent backgrounds, VFX style, and visual quality still depend on prompts, models, and workflow quality.