Workflows
Advanced ComfyUI workflows ready to download and use.
MiniMax H3 in ComfyUI: Motion Director for Longer Videos on RTX 3090
We tested Motion Director on RTX 3090 to get past MiniMax H3's length limit: 3 chained 5s segments in one 15s clip. Pure I2V vs. Ref2VA with a character sheet.
MiniMax H3 in ComfyUI: Real Face Swap with Ref2VA on RTX 3090
We verified MiniMax H3 face swap (RTX 3090) twice: with the creator's own assets and with a 100% self-generated video. The title's 'body swap' isn't shown.
MiniMax H3 in ComfyUI: PDD Acc LoRA (8-Step) vs Turbo (4-Step) on RTX 3090
Alibaba PAI released an 8-step LoRA for MiniMax H3. We tested it on RTX 3090: the 4-step turbo doesn't look bad, it just doesn't advance the choreography over time.
MiniMax H3 in ComfyUI: Sage Attention, Sol-Attn, Latent Upscale on RTX 3090
We tested MiniMax H3 speed tricks on an RTX 3090: Sage Attention, Kijai Sol-Attn, two-stage latent upscale. One hid a real bug no tutorial mentions.
Uni3C ControlNet: Real Camera Control for Wan 2.1 in ComfyUI
I tested native Uni3C ControlNet in ComfyUI on an RTX 3090: point-cloud camera rendering in Docker plus Wan 2.1 generation. Two real blockers fixed.
COMFYCLAW: We Tested the Agent That Builds and Repairs ComfyUI Workflows
We tested COMFYCLAW, the research agent that edits ComfyUI graphs and self-corrects with a visual verifier, using Claude Code as backend on an RTX 3090.
Camera Movement Transfer: IC-LoRA Cameraman v2 + LTX 2.3 in ComfyUI
I tested IC-LoRA Cameraman v2 on LTX 2.3: copies camera movement from a reference video onto a new one, without touching content. Workflow and OOM bug documented.
HunyuanVideo 1.5 in ComfyUI: Real RTX 3090 Install and Test Guide (I2V 480p)
How to install and run HunyuanVideo 1.5 (8.3B, Tencent, open-source) in ComfyUI: exact models, a step-by-step workflow for the I2V 480p step-distilled variant, and the real generation time measured on an RTX 3090.
ComfyUI Upscale Workflow: Ultimate SD Upscale Setup (Free JSON Download)
Upscale ComfyUI images to 4K without running out of VRAM. Ultimate SD Upscale, ESRGAN models, and tiled VAE decode explained -- with a downloadable workflow.
5 Local AI Video Tests on One RTX 3090: What Actually Worked in 2026
LTXV-2.3 distilled, LTXV-2.3 dev, SCAIL-2, Wan 2.1 I2V, and Wan 2.2 I2V, all tested for real on the same hardware. Real timings, real crashes, real quality differences -- not a spec-sheet comparison.
LTX Director in ComfyUI: I Tested the New Timeline Node and Found 2 Real Bugs
Tested the community LTX Director node on top of LTXV-2.3 distilled on an RTX 3090: a version-compatibility bug and an empty-prompt bug, both documented and fixed, plus the real generated video.
I Maxed Out an RTX 3090 with LTX-2.3's Dev Model — It Crashed 4 Times
A reproducible crash 4/4 times running LTXV-2.3's dev model on a 24GB RTX 3090: the bottleneck wasn't VRAM, it was system RAM via the GGUF loader path.
Viral LTXV-2.3 + RTX Super Resolution Workflow in ComfyUI, Reproduced
I reproduce a viral text-to-video-with-audio workflow using LTXV-2.3 chained into RTX Video Super Resolution. Real times, VRAM, and gotchas on an RTX 3090.
SCAIL-2 in ComfyUI: Real Character Replacement Test (No Custom Nodes, RTX 3090)
Tested Comfy-Org's official SCAIL-2 template for AI character replacement: swap a person in any video for a reference image while keeping the scene and motion. Real models, real fixes, real output video.
I Animated the Same Frame with Wan 2.1 I2V That LTXV-2.3 Generated From Scratch
Took the exact first frame from a previous LTXV-2.3 text-to-video test and animated it with Wan 2.1 I2V instead -- a real, controlled comparison of two different local video generation approaches on the same RTX 3090.
Wan 2.2 I2V vs Wan 2.1: Same Frame, Same RTX 3090, Different Quality
Ran the same image-to-video test with Wan 2.2's dual-model MoE architecture instead of Wan 2.1: faster and visibly sharper. Real numbers from a local RTX 3090.
Wan 2.2 Video Generation in ComfyUI: Image-to-Video Workflow Guide
Animate any image in ComfyUI with Wan 2.2. Which model size to pick for your GPU, node setup, and motion prompts that actually work.