Two learning tracks for running H3 and creating videos, with in-depth selections, source resources, requirements, and known issues.
- ComfyUI zero to your first H3 video with audio
Start with a clean ComfyUI setup, place all five model files correctly, and render a playable T2VA MP4 with picture and synchronized audio.
- Turbo + lower VRAM: reduce 20 steps to 4–8
Add the community Turbo nodes to a working H3 ComfyUI baseline, render usable clips at 6–8 steps, and follow a controlled fallback order when VRAM runs out.
- Run H3 on Mac from zero with pure C + Metal
Without Python, PyTorch, or ComfyUI, compile h3.c on Apple Silicon, validate the model, and generate a first MP4 with audio.
- Motion Context: chain H3 clips into a longer video
Install Motion Context, carry the previous audio-video latent into the next generation, and complete a two-clip chain with continuous picture and sound.
- From a clean machine to the first video with audio
Build a reproducible local H3 setup and prompt-writing baseline.
- How four essential H3 repositories fit together
Place the official Skill, Turbo, Motion Context, and audio workflows into one clear path.
- Attention backends, character LoRA, and low-VRAM paths
Add steadier settings, an attention backend, character LoRA, or a low-VRAM path based on the actual need.
- H3 prompts: plan shots, dialogue and sound
Choose the generation mode, understand an official prompt example, and check the alignment of shots, dialogue and sound.
- Character references and dialogue: assign image, video and audio roles
Use Ref2VA to assign identity and sound references, then review dialogue and identity retention before trying camera edits.
- Turn storyboard frames into a consistent trailer
Generate shot by shot from final-film frames and shot descriptions, then assemble a trailer.
- Build a one-minute trailer on RTX 3060 with Turbo LoRA
Animate an existing image set into short shots and edit them into a one-minute trailer.
- Maestro six-step Turbo: trade detail for iteration speed
Use a six-step distilled LoRA in Maestro for fast H3 previews.
- Low-VRAM Maestro: pruned 20B plus Turbo
Reduce VRAM pressure with the pruned 20B model, six-step Turbo, and optional First Block Cache.
- Before-and-after Turbo LoRA benchmark on RTX 3060
Compare native and Turbo runs at fixed resolution and duration on the same RTX 3060.
- Run H3 Turbo on Mac Studio with Phosphene
Use fewer steps on Apple Silicon to reduce a five-second run from about 20 minutes to about eight in the reported setup.
- H3 in the cloud: your first video without local setup
Open a cloud template, replace its prompt or first frame, then generate and download an MP4 with audio.
- Dual-clock sampling: separate video and audio step counts
Control video and audio sampling separately in ComfyUI and compare it carefully with an acceleration LoRA.
- Image to video: first frames, last frames and modes
Animate a still image or constrain the opening and ending with two images.
- RTX 4080 16GB: pruned plus INT8 low-VRAM setup
Run a ten-second local H3 test within 16GB VRAM.
- 4-bit plus DiffSynth: an 8GB-VRAM route
Run H3 on smaller GPUs through 4-bit quantization and VRAM-adaptive inference.
- H3 Ref2VA LoRA: train a video LoRA with AI Toolkit
Explore the author’s Ref2VA LoRA training route and identify its tool, dataset format and hardware requirements.
- H3 WebUI: Motion Context plus built-in upscaling
Chain short clips toward roughly one minute with Motion Context and upscale before export.
- 4070 12GB: chain three roughly 5-second chunks into a 13-second character dance
When a 13-second pass does not fit, use eight-step Turbo and Motion Context chunks.
- Four-step vs eight-step LoRA: a 30-second chaining A/B
Compare lightx2v and larryvrh LoRAs for sharpness, noise, speech, and stability.
- MiniMax Director: multi-shot video with sound on a timeline
Arrange shots, recurring subjects and dialogue in ComfyUI, inspect the compiled prompt, and render an MP4 with sound.