41 Free AI Workflows & Guides

MiniMax H3 Audio Lip Sync

Turn a single photo into a video of that person singing or speaking your audio — with lip movement that matches almost perfectly. This is a MiniMax H3 image-to-video workflow I put together for audio-driven lip sync: feed it a…

MiniMax Music 3 — Local AI Song Generation

MiniMax Music 3 is a new open weights AI music generation model, and it's one of the strongest local alternatives to Suno yet. Give it a set of lyrics and a description of the sound you want, and it generates a complete,…

MiniMax H3 Turbo LoRAs (4–8 Step)

A follow-up to my MiniMax H3 workflow: there are now two turbo LoRAs that cut sampling from ~20 steps down to 4-8 — roughly a 4x speedup — while keeping H3's native audio. This post covers both, with attached workflows for each…

MiniMax H3 — Local AI Video with Native Audio

MiniMax H3 is the first open weights model in MiniMax's Hailuo video line, and it's a big one — a general purpose, omni-modal generator that understands text, images, video, and audio together and produces video with native…

Krea 2 Identity Image Edit LoRA

The Krea 2 Identity Image Edit LoRA brings instruction based, identity preserving image editing to Krea 2 in ComfyUI. You give it an image and a plain language instruction — "create a photo of this person at a night market" — and…

LTX 2.3 Outpainting LoRA

This is a ComfyUI workflow built to run the LTX-2.3 22B model alongside the IC-LoRA Outpaint module created by oumoumad. It allows you to expand and outpaint both still images and existing video clips.

WAN 2.1 Low VRAM Text-to-Image

Generate high-resolution 1024 × 1024 images in under 30 seconds on just 6 GB VRAM using the WAN 2.1 video generation model.

Hunyuan3D 2.1 — Image to 3D

Create high-resolution 3D mesh assets in ComfyUI with Hunyuan 3D 2.1, a major upgrade from version 2.0. The new model delivers significantly improved asset quality, featuring smoother mesh surfaces, more captured detail from the…

Flux.1 Kontext Dev GGUF

I'm excited to share that the Flux.1 Kontext Dev GGUF model is now available on Hugging Face. I've created a simple, free custom low VRAM workflow to help you get started with the model using as little as 4GB of VRAM.

HeartMuLa Song Generator

Open-source, Apache 2.0 AI music generation that writes full songs with sung lyrics from a style prompt — the closest thing to running Suno locally. Includes a lyric transcribe path and merged single-file models.

LTX 2.3 Director v2 (12 GB GGUF)

Timeline-style video editing inside ComfyUI. Drop images on a track, layer prompts across segments, add custom audio and IC-LoRA control — replacing separate t2v, i2v and multi-keyframe workflows with one node. GGUF quantised to fit 12 GB.

Krea 2 Image-to-Image Depth ControlNet LoRA

Depth-guided image-to-image in ComfyUI that preserves 3D structure and composition while letting you restyle content via prompts. Works with Krea-2 Raw and Turbo.

Krea 2 Turbo Text-to-Image

Free, high-performance open-source text-to-image model tuned for local ComfyUI generation. Supports FP8 and GGUF for lower VRAM with fast, high-quality output.

Ideogram 4.0 Text-to-Image

ComfyUI workflow with strong prompt adherence and best-in-class typography — great for posters, logos, mockups and social graphics.

Flux 2 Klein 9B Upscale + Adonis Refiner LoRA

Upscale images while keeping consistency and adding refined detail using Flux 2 Klein 9B and the Adonis Refiner LoRA. GGUF build lowers VRAM use.

Flux 2 Klein 9B Character FaceSwap

Face/head swap workflow combining Flux 2 Klein 9B Edit with a BFS character face-swap LoRA — runs on modest 6GB-VRAM GPUs.

LTX-2.3 Audio-to-Video

Turn a still image + audio into an animated clip with accurate lip-sync in ComfyUI — build talking avatars and singing characters.

Anima 2B Anime Text-to-Image

A 2B-parameter anime-optimized diffusion model with clean linework and expressive color. Run locally in ComfyUI or via free Colab.

LTX-2 Audio-to-Video (Kijai)

Convert spoken audio into synchronized talking-avatar video, tuned for limited-VRAM systems while keeping quality high.

SmoothMIX WAN 2.2 Finetune

WAN 2.2 finetune for fast, smooth image-to-video — 5-second 480p clips in 4–6 steps with no LoRA training required.

Jib Mix ZIT — Z-Image Turbo Finetune

A Z-Image Turbo finetune tuned for cleaner, more realistic portraits with sharper detail and smooth skin — fast, low-step, consumer-GPU friendly.

Meta AI SAM 3 — Segment Anything

Object detection, segmentation and tracking across images and video with Meta's SAM 3. Includes three ready ComfyUI workflows (boxes, points, video tracking).

SeedVR2 AI Video Upscaling

ByteDance's SeedVR2 restores resolution with strong temporal consistency. 480p upscales in 3–10 min on a 6GB RTX 4050; standard + GGUF workflows.

Microsoft VibeVoice TTS

Run Microsoft's VibeVoice text-to-speech locally on Windows via Gradio or ComfyUI, with custom voice cloning support.

Qwen Image Edit — Fast 4-Step + LoRA

Qwen Image Edit with 4-Step Lightning LoRA and upscaling — high-quality semantic edits and object manipulation in under 90s on 6GB VRAM.

Qwen-Image 4-Step Text/Image-to-Image

Alibaba's 20B Qwen-Image with Lightning LoRA acceleration — text-to-image, image-to-image and experimental inpainting in ~80s on mid-range GPUs.

Flux Krea Dev Nunchaku Text-to-Image

Quantized Flux Krea that generates 1024×1024 photoreal images in under a minute on consumer GPUs using MIT Han Lab's Nunchaku acceleration.

WAN 2.2 14B Low-VRAM Text/Image-to-Video

State-of-the-art local video generation on as little as 6GB VRAM in ComfyUI. 480p clips in ~10–15 min with quantized GGUF models.

WAN 2.2 14B Low-VRAM Text-to-Image

High-resolution image generation on 6GB VRAM using quantized GGUF WAN 2.2 — cinematic-quality images with fewer artifacts.

Nunchaku Flux Dev + Kontext

Accelerated Flux Dev and Flux Kontext workflows using MIT's Nunchaku framework — up to 2× faster generation while keeping quality on consumer GPUs.

OmniGen2 Multimodal Image Editor

Open-source model for text-to-image, image editing and multi-image composition — runs efficiently on 6GB+ VRAM in ComfyUI.

Ace-Step AI Music Generator (Colab)

An improved Google Colab notebook for the open-source Ace-Step music generation model, fixing issues in the original implementation.

3D Asset Editing & Inpainting — Trellis Hi3DGen (Colab)

Free Colab notebook for 3D mesh editing and inpainting: upload .ply meshes + conditioning images to perform guided 3D-to-3D transformations.

Flux Colab Notebooks — ControlNet, Redux, Inpaint, Outpaint

Four free Colab notebooks (by LucipherDev) to run Flux ControlNet, Redux, inpainting and outpainting in Google Colab at no cost.

Stable Diffusion Forge + Flux in Google Colab

Run Flux Dev NF4 with Stable Diffusion Forge UI in Google Colab — LoRA support plus Google Drive integration for models and outputs.

WAN 2.1 Self Forcing — Text-to-Video + VACE

WAN 2.1 Self Forcing streams 480p video in near real-time with 150–400× lower latency — fast text-to-video and VACE image-to-video on as little as 6GB VRAM.

WAN SCAIL 2 — Animation Transfer (Video-to-Video)

WAN SCAIL 2 video-to-video animation transfer with smoother motion, better temporal coherence and identity retention; the GGUF build runs on 16GB+ VRAM.

WAN 2.2 SVI 2.0 — Infinite-Length Video

WAN 2.2 SVI 2.0 creates seamless, infinite-length video in ComfyUI using Stable Video Infinity LoRAs, tuned for low VRAM and faster rendering.

Z-Anime — Anime Finetune of Z-Image

Z-Anime is an anime-focused finetune of the Z-Image base for high-quality anime generation on 6GB VRAM — 1024×1536 images in under 2 minutes.

Z-Image Turbo Character FaceSwap

Character face-swapping with Z-Image Turbo + Meta SAM 3 — low-VRAM image-to-image editing controlled by denoise, CFG scale and LoRA strength.

Trellis 2 — Microsoft SOTA 3D Asset Generation

Microsoft's Trellis 2 generates high-quality 3D assets in ComfyUI with sharper contours and better texture mapping — under 2 minutes on an RTX 4090.