AI Video Generation Guides
Long-form notes and measured write-ups from this site. The tools live in the nav above; everything that is a read, not a calculator, is here.
SeedVR2 ComfyUI: Native Nodes or the numz Node, Files and VRAM
Two ways to run SeedVR2 in ComfyUI: built-in nodes since v0.28.0, or the numz upscaler node. Every model file with exact sizes, block swap, tiling and 4n+1.
ComfyUI Frame Interpolation: RIFE vs FILM, Models and FPS Math
ComfyUI frame interpolation today: the built-in RIFE and FILM nodes, the Fannovel16 node pack, exact model files, the fps math for Wan, and the common errors.
Wan 2.2 First Last Frame in ComfyUI: Which Model and Which Files
Wan 2.2 first last frame in ComfyUI: no dedicated FLF2V model, the I2V-A14B experts plus one node, file sizes, template defaults and the Wan 2.1 FLF2V model.
ComfyUI Missing Nodes: How to Install Them and Why They Stay Red
What the ComfyUI missing nodes warning means, how to install packs with Manager or by hand, and why nodes stay red: import errors, Python, settings.
Wan2GP vs ComfyUI: Which to Use for Local AI Video on a Small GPU
Wan2GP vs ComfyUI from their own docs: supported models, MiniMax H3, memory profiles, installs, model downloads, LoRAs, update pace and the two licences.
ComfyUI TeaCache: Import Failed, Wan 2.2 and What Works Now
Why ComfyUI-TeaCache shows Import Failed on current ComfyUI, why it has no Wan 2.2 option, and which TeaCache, MagCache and EasyCache nodes still work.
Wan 2.2 VACE: Which Model It Is, Its Files and ComfyUI Workflow
Wan 2.2 VACE is Alibaba PAI's Wan2.2-VACE-Fun-A14B, not a Wan release. Exact file sizes, GGUF quants, the native WanVaceToVideo route and what it does.
Wan 2.2 S2V: Files, ComfyUI Workflow and Clip Length
What Wan2.2-S2V-14B is, every file with its exact size, the wav2vec2 audio encoder, the native ComfyUI template and nodes, and how longer clips are built.
Wan 2.2 Text to Image: Single-Frame Stills in ComfyUI
How Wan 2.2 text to image works in ComfyUI: one frame from the T2V template, low-noise-only vs both experts, exact file sizes, LoRAs and what is official.
FramePack VRAM Requirements: 6GB Minimum and High-VRAM Mode
FramePack VRAM requirements from the source code: the 6 GB minimum, the 60 GiB high-VRAM switch, the preservation slider, download sizes and ComfyUI routes.
ComfyUI extra_model_paths.yaml: Location, Example and Fixes
Where ComfyUI extra_model_paths.yaml lives for portable, manual and Desktop installs, a working Windows example, every valid key, and why it does not load.
ComfyUI Video Combine Node: Every Setting, CRF and Loop Count
Every input of the Video Combine node with its default, what crf does per format, loop_count, pingpong, ffmpeg errors, and when native Save Video is enough.
Wan 2.2 VRAM Requirements: 5B vs 14B, 8GB to 24GB
Wan 2.2 VRAM requirements by model and card: exact file sizes for 5B and 14B, what the official numbers measure, and a starting point for 8, 12, 16 and 24 GB.
Wan 2.2 ComfyUI Models Folder: Every File and Its Path
Which ComfyUI models folder each Wan 2.2 file goes in, for the 5B and every 14B template, with exact sizes, the GGUF folder and four mistakes in the official docs.
LTX Video on 12GB VRAM: The Workflow, the Files and the One Measured Run
Lightricks says LTX needs 32GB of VRAM. One published run did LTX-2.5 on a 12GB card with 48GB of RAM. The workflow, exact file sizes, GGUF options and settings.
ComfyUI Black Video Output: Causes and Fixes for Wan, LTX and HunyuanVideo
Why ComfyUI saves an all-black video, how to tell which stage broke, and the fixes maintainers have confirmed: SageAttention, VAE precision, tiled decode and wrong VAEs.
ComfyUI diffusion_models vs checkpoints: Which Folder Your Model Goes In
The difference between a checkpoint and a diffusion model file in ComfyUI, which loader reads which folder, what happens when a file is in the wrong one, and where GGUF goes.
RTX 3060 12GB AI Video Models: What Runs, What Fits, What Was Measured
Which open AI video models run on an RTX 3060 12GB in ComfyUI: Wan, HunyuanVideo 1.5, LTX, CogVideoX, Mochi and MiniMax H3, with file sizes, official figures and the system RAM they need.
MiniMax H3 Singularity Is a Ref2VA Fine-Tune: Files and Licence
MiniMax H3 Singularity is a community Ref2VA fine-tune of H3: exact file sizes, the ComfyUI folder for each, the workflow it expects, memory and licence.
MiniMax H3 NVFP4, 40% Smaller: Files, GPUs and ComfyUI
MiniMax H3 NVFP4 runs natively only on Blackwell, such as RTX 50-series cards. Exact file sizes, ComfyUI folders, and what RTX 30 and 40 cards do instead.
MiniMax H3 GGUF Starts at 6.7 GB: Quants, Loaders, VRAM
MiniMax H3 GGUF files for ComfyUI, 6.7 to 36 GB: exact sizes by quant and repo, which loader reads which file, and what fits a 12, 16 or 24 GB card.
HunyuanVideo 1.5 VRAM Requirements: 480p, 720p, 8GB to 24GB
HunyuanVideo 1.5 VRAM requirements by file and card: exact sizes for fp16, fp8 and GGUF, what the 14 GB minimum measures, and a starting point for 8 to 24 GB.
Wan 2.5 Is Not Open Source: Weights, API and Local Options
Wan 2.5 has no public weights. What Alibaba released under Apache 2.0, where Wan 2.5 runs as a paid API, and the closest model you can run locally.
ComfyUI Out of Memory? Fix by Stage: KSampler, VAE Decode, RAM
Match ComfyUI's out-of-memory message to the stage that failed, then fix it: flags checked in ComfyUI's source, tiled decode, fp8 files, swap and pagefile.
Wan 2.2 GGUF: Which Files, Which Quant and How to Load Them
Wan 2.2 GGUF in ComfyUI: the QuantStack and bullerwins repos, exact sizes from Q2_K to Q8_0 for both 14B experts, the GGUF loader, text encoder and speed caveats.
Wan 2.2 Lightx2v LoRA: Versions, Settings and Slow Motion
Every Wan 2.2 lightx2v LoRA release with exact sizes and dates, the 4-step settings ComfyUI templates use, and what lightx2v says about the slow-motion problem.
Wan 2.2 Animate in ComfyUI: Workflow, Files and Custom Nodes
Wan 2.2 Animate in ComfyUI: the native template, every file it loads with exact sizes, the three custom nodes it needs, GGUF builds and the Kijai wrapper route.
LTX 2.3 vs Wan 2.2: Specs, File Sizes, Licence and VRAM
LTX 2.3 vs Wan 2.2 from primary sources: parameters, audio, exact Hugging Face file sizes, official VRAM statements, ComfyUI routes and the two licences.
SageAttention ComfyUI: Install on Portable, Desktop and Linux
Install SageAttention for ComfyUI on Windows portable, Desktop or Linux, turn it on with --use-sage-attention or a node, and check the log line that proves it.
Torch Not Compiled with CUDA Enabled in ComfyUI: Fix by Install
Why ComfyUI says Torch not compiled with CUDA enabled, how to check your PyTorch build, and the fix for portable, Desktop, venv, AMD, Mac and Stability Matrix.
MiniMax H3 Recast: How to Use It, Limits and Pricing
How to use MiniMax H3 Recast to swap people in a 5–30 s clip from up to 4 photos: step-by-step, API call, exact limits, pricing and a local ComfyUI route.
Qwen Image 2.1 License: Commercial Use, Outputs
The Qwen Image 2.1 license is research-only, not Apache 2.0. See what it permits, what Qwen said about outputs and what needs a separate licence.
MiniMax H3 ControlNet Union 2.0: ComfyUI Setup
MiniMax H3 ControlNet Union 2.0: eight control types plus inpainting, native in ComfyUI 0.38.0. Exact file sizes, folders, node names and VRAM notes.
FastH3 VRAM Requirements: 8-Step V2 Sizes by Build
FastH3 VRAM requirements have no official consumer figure. Exact FastH3 8-Step V2 file sizes for BF16, INT8 and GGUF, full-set totals and what each card holds.
Qwen Image 2.1 VRAM Requirements: Sizes by Build
Qwen Image 2.1 VRAM requirements have no official minimum. See exact file sizes per build, full-set totals and a starting point for your card.
MiniMax H3 4× Faster: Turbo LoRA, Cache & Sage Attention
4.1× faster on an RTX 3060, same seed, with free files — no paid API. Which Turbo LoRA to load, which cache nodes stack with it, where Sage Attention fits.
MiniMax H3 Max by fal: API, Pricing, Speed & Limits
Independent guide to fal Research’s post-trained MiniMax H3 Max: API endpoints, standard pricing, speed evidence, limits, and when to use standard H3.
MiniMax H3 on RTX 3060 12GB: Our Measured 43-Minute Run
MiniMax H3 on an RTX 3060 12GB, measured twice: 43 minutes, 94.8% VRAM, 43GB system RAM. Full test card, hashes and the hardware failure we hit.
video_minimax_h3_t2v.json: Files, VRAM and Run Cost
Download video_minimax_h3_t2v.json and inspect a 2,175.5-second 1344×768 run on an RTX 3060 12GB, with required files, VRAM, RAM and canvas cost.
MiniMax H3 Cache Nodes Tested on One RTX 3060
Eight MiniMax H3 cache nodes for ComfyUI, installed and measured on the same RTX 3060: which ones load, which ones change the output, and how much time each saves.