- Home
- Skills
- Image & Video
- Local ComfyUI Image & Video Gen (8GB VRAM)
More screenshots
Works with the AI Agents you already use
Local ComfyUI Image & Video Gen (8GB VRAM)
Run SDXL, Krea2, Qwen-Image 2.1, Wan 2.2 & MiniMax H3 locally via ComfyUI—tuned for consumer 8GB VRAM + 32GB RAM setups with zero cloud API fees.
$10
Local ComfyUI Image & Video Gen (8GB VRAM)
Example session with this skill installed
Use minimax_h3_w4a8_t2av to generate a 5-second 720p talking character video on my 8GB VRAM + 32GB RAM machine: a silver-haired anime presenter in a cozy studio saying "Welcome to my local AI studio, running 100% offline!"
- Read your context and instructions
- Compiled the local comfyui image
Generated the synchronized audio-video clip locally using the MiniMax H3 W4A8 workflow with CPU offloading enabled for 8GB VRAM + 32GB RAM.
Generation Summary
| Field | Value |
|---|---|
| Workflow | minimax_h3_w4a8_t2av.json |
| Server | http://127.0.0.1:8188 (auto-verified via ensure_server.py) |
| Prompt | Silver-haired anime presenter in a cozy studio, synchronized English speech |
| Quantization / Offload | W4A8 + 32GB RAM CPU Offload (Peak VRAM: ~7.4 GB) |
| Output Path | ./outputs/minimax_h3_0001.mp4 |
| Closeout Check | Passed (closeout_check.py exit code 0) |
Next steps
- Open
./outputs/minimax_h3_0001.mp4to verify lip-sync and audio clarity. - Want to lock character appearance across multiple shots? Re-run using
minimax_h3_w4a8_i2vwith your frame as the reference image.
Connects securely to your tools. The creator never sees your data.
About this skill
Running high-end image and video generation locally usually feels impossible without a 24GB flagship GPU—forcing you into expensive cloud APIs. This skill provides a pure Python bridge to your local ComfyUI instance, packed with quantized and offloaded workflows specifically tuned to run smoothly on consumer 8GB VRAM + 32GB RAM hardware with zero external API keys or cloud costs.
What it does
- 8GB VRAM + 32GB RAM Ready: Pre-configured with GGUF/W4A8 quantization, BlockSwap, CPU offloading, and memory-safe defaults so consumer GPUs can run heavy pipelines without OOM crashes.
- Multi-Model Image Generation: Orchestrates SDXL Illustrious and Krea2 (12.8B semi-realistic/photographic T2I via
Krea2BlockSwap) for high-fidelity generation. - Advanced Image Editing & Identity Lock: Performs instruction-based editing, outpainting, inpainting, and face-swap with Qwen-Image 2.1 Turbo, plus identity-consistent character editing via Krea2 Identity Edit.
- Local Video Synthesis: Runs heavy DiT video models including Wan 2.2 (4-step I2V) and MiniMax H3 (W4A8 T2AV & I2V) on local GPUs.
- Character Animation & Motion Transfer: Implements character animation and replacement via Wan-Animate-2 and SCAIL-2, plus H3's first-frame character locking.
- Audio Synchronization: Generates video with synchronized speech and lip-sync using the MiniMax H3 audio-video branch.
How it works
- Server Management: Uses
ensure_server.pyto check for an active ComfyUI instance or launch it headlessly in the background. - Workflow Selection: Picks a pre-configured API-format JSON workflow from the
assets/directory. - Parameter Fusion: Merges your prompt and parameters into the workflow using declarative
_tp.jsontemplates. - Execution & Polling: Submits the task to the local API, polls the queue until completion, and downloads the outputs.
- Validation: Every run concludes with a mandatory closeout check to ensure output integrity and log consistency.
Frameworks & tools
Built with Python 3.10+ standard library (zero external pip install dependencies). Interfaces with a local ComfyUI server (v0.3.x+ / ComfyUI Desktop recommended) on http://127.0.0.1:8188. Supports GGUF, W4A8, and Safetensors model formats.
Why this beats prompting it yourself
Raw prompting cannot handle the multi-stage JSON structure required by the ComfyUI API or manage tight 8GB VRAM constraints like model offloading, BlockSwap, and cache clearing. This skill automates server lifecycle, parameter injection, and artifact retrieval so you can focus on the creative output.
Use cases
- Automated Asset Pipelines: Generate variations of character art or environment backdrops on a standard 8GB GPU workstation or laptop.
- Virtual Presenters: Create short video clips of characters speaking specific lines with synchronized audio.
- Image Retouching: Programmatically swap outfits, faces, or backgrounds in existing photos using natural language.
- Motion Transfer: Apply the motion from a driving video to a static character reference.
Requirements & known limitations
- Recommended Hardware: Local NVIDIA GPU with 8GB+ VRAM and 32GB+ system RAM (for CPU offloading of larger video/DiT models).
- Local Setup: Requires local ComfyUI (v0.3.x+) running on
http://127.0.0.1:8188. Model weights, LoRAs, and custom nodes (such asComfyUI-Krea2BlockSwap) are not bundled in this package and must be installed in your local ComfyUI environment. - Format: Only works with workflows exported in ComfyUI API format.
Free lifetime updates included as new local ComfyUI workflows and models are added.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Video Skills.
Local GPU with 8GB+ VRAM required. ComfyUI must be in API mode. Some models require specific custom nodes like Krea2BlockSwap. No support for cloud APIs.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 4 days ago
- Software
Python 3.10+ComfyUI (v0.3.x+)NVIDIA GPU (8GB+ VRAM, 32GB+ RAM)
Git
- Optional
- ComfyUI-Krea2BlockSwapComfyUI-VideoHelperSuite
Connects strictly to your local ComfyUI server on http://127.0.0.1:8188 (zero external API calls). Model weights and LoRAs are not included in this ZIP and must be placed in your local ComfyUI models folder.
Permissions
Network access is used strictly for local loopback HTTP requests to the user's ComfyUI server (http://127.0.0.1:8188), with zero external network calls. Environment variables are used only for optional local port/directory configuration.