# AI use cases for Mini Max H3 Director Cut Studio (karuvanan/MiniMax-H3-Director-Cut-Studio) Repo: [karuvanan/MiniMax-H3-Director-Cut-Studio](https://github.com/karuvanan/MiniMax-H3-Director-Cut-Studio) | Use cases: 4 | Page: https://dedupedai.com/use-cases/karuvanan-minimax-h3-director-cut-studio/ A local PySide6 director workbench that plans multi-track AI video shots and submits them to ComfyUI for MiniMax H3 Ref2VA rendering. This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL. - Plan a multi-shot AI video on a timeline: https://dedupedai.com/use-cases/karuvanan-minimax-h3-director-cut-studio/plan-multi-shot-ai-video-timeline/ - Swap a cast member in an existing source video: https://dedupedai.com/use-cases/karuvanan-minimax-h3-director-cut-studio/swap-cast-member-in-source-video/ - Generate Design concept images with Z-Image and LM Studio: https://dedupedai.com/use-cases/karuvanan-minimax-h3-director-cut-studio/generate-design-concept-images-with-lm-studio/ - Split a song into vocal and music tracks locally: https://dedupedai.com/use-cases/karuvanan-minimax-h3-director-cut-studio/separate-vocals-and-music-with-kim-vocal-2/ ## How to plan a multi-shot AI video on a timeline with Mini Max H3 Director Cut Studio Repo: [karuvanan/MiniMax-H3-Director-Cut-Studio](https://github.com/karuvanan/MiniMax-H3-Director-Cut-Studio) | Difficulty: Advanced | Time: 4 hours | Basis: From the README Lay out Shots, Dialogue, Markers and Prompts on a multi-track timeline, then submit the valid media to ComfyUI. **Who it is for:** A solo filmmaker or creator with a local GPU and ComfyUI who wants to turn a script or image set into a short AI-rendered sequence. **The problem:** Writing separate prompts and rendering clips one at a time makes it hard to keep shots, dialogue and timing consistent across a sequence. **The outcome:** A planned timeline of shots and dialogue with generated MiniMax H3 Ref2VA prompts, rendered through ComfyUI using the default workflow. ### Steps 1. Complete Step 1 in the README (step1.md) on the machine that will run the studio. 2. Download the model files for `video_minimax_h3_r2v_9image_3audio_3video_api.json` (UNETLoader, CLIPLoader, two VAELoaders and the H3 Turbo LoRA) and place them in the ComfyUI folders the README lists. 3. Start the studio from the project root with `.\run_h3_prompt_studio.bat`. 4. Add images, video and audio to the Virtual Media Pool, then build Shot, Dialogue, Marker and Ending Hold items on the Timeline. 5. Generate the MiniMax H3 Ref2VA prompts and submit the current valid media and parameters to ComfyUI. ### Prompt to paste into your agent ```text Using the repo karuvanan/MiniMax-H3-Director-Cut-Studio, help me place the five H3 model files for video_minimax_h3_r2v_9image_3audio_3video_api.json into the ComfyUI folders named in the README, then build a three-shot timeline for . Expected output: a list of the generated shot prompts. Verify that each model file name matches the README exactly before submitting to ComfyUI. ``` ### Caveat The README says model weights are not shipped with the source code, so you must download and name every listed model file yourself before the default workflow will run. Risk, voice cloning: Dialogue can be voiced with Qwen3-TTS, VoxCPM2 or Edge TTS, so confirm you have rights to any voice you clone or reproduce. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/karuvanan-minimax-h3-director-cut-studio/plan-multi-shot-ai-video-timeline/ ## How to swap a cast member in an existing source video with Mini Max H3 Director Cut Studio Repo: [karuvanan/MiniMax-H3-Director-Cut-Studio](https://github.com/karuvanan/MiniMax-H3-Director-Cut-Studio) | Difficulty: Advanced | Time: 4 hours | Basis: From the README Replace a person in a source video shot by shot using the Character-Swap LoRA, the H3 Relay and SAM3.1 masks. **Who it is for:** A video editor or creator who has a source clip and needs to replace one actor with a reference image, and has the Character-Swap models installed. **The problem:** Re-shooting a scene to change one performer is costly, and simple masking leaves ghosting or edge artifacts when the new person has a different body or pose. **The outcome:** A shot-level cast remake of the source video where the selected roles are swapped and the rest of the frame is kept, with dirty segments regenerated. ### Steps 1. Install the `akatz-ai/h3-relay` custom node into `ComfyUI/custom_nodes/h3-relay/` and place every Character-Swap model listed in the README, including `sam3.1_multiplex_fp16.safetensors` for multi-character swaps. 2. Confirm ffmpeg and ffprobe run from the PATH on the server. 3. Optionally run `install_h3_cache_status_server.bat` on the ComfyUI host to show cache, VRAM and GPU status in Studio and the Windows Tray. (inferred, check the README) 4. Start the studio with `.\run_h3_prompt_studio.bat`, choose `Source Video Cast Remake H3`, and open the Shot Visual Editor. 5. Check the roles to swap, choose a P Picture for each, leave other people on `KEEP V1`, click `SAVE SHOT`, then click `REGENERATE DIRTY`. ### Prompt to paste into your agent ```text Using the repo karuvanan/MiniMax-H3-Director-Cut-Studio, set up the Source Video Cast Remake H3 path for the source clip and swap the role named with the reference image

. Expected output: a regenerated shot in the studio with the swap applied to that role only. Verify that LabsH3PrepareVideo, H3RelayWindowedEdit, SAM3_VideoTrack and SAM3_TrackToMask appear in ComfyUI without IMPORT FAILED. ``` ### Caveat The README lists many separate model files, a custom node and system ffmpeg that must all be in place before the Character-Swap path runs, and the source footage license must be checked separately. Risk, voice cloning: The swap replaces a real person's appearance in footage, so you need permission from the performer and the source video owner before using the output. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/karuvanan-minimax-h3-director-cut-studio/swap-cast-member-in-source-video/ ## How to generate Design concept images with Z-Image and LM Studio with Mini Max H3 Director Cut Studio Repo: [karuvanan/MiniMax-H3-Director-Cut-Studio](https://github.com/karuvanan/MiniMax-H3-Director-Cut-Studio) | Difficulty: Intermediate | Time: 2 hours | Basis: From the README Create concept reference images with the Z-Image workflow and write prompts through a local LM Studio model. **Who it is for:** A creator or art director preparing reference frames for a video project who wants local image generation and a local text model. **The problem:** Producing consistent concept images and matching text prompts by hand takes time, and the two steps are hard to keep in sync. **The outcome:** Concept reference images from the Design page, using the Z-Image text-to-image workflow, with prompts generated through the LM Studio OpenAI-compatible API. ### Steps 1. Place `z_image_turbo_bf16.safetensors`, `qwen_3_4b.safetensors` and `ae.safetensors` in the ComfyUI folders named in the README. 2. Load a text model in LM Studio and start its local server at `http://127.0.0.1:1234/v1`. 3. Start the studio with `.\run_h3_prompt_studio.bat` and open the Design page. 4. Enter the LM Studio Base URL and pick a Model ID from the live list returned by `/v1/models`, then click `TEST CONNECTION`. 5. Generate the concept images and review the captions the local BLIP model creates for them. ### Prompt to paste into your agent ```text Using the repo karuvanan/MiniMax-H3-Director-Cut-Studio, configure the Design page to use the LM Studio Base URL http://127.0.0.1:1234/v1 and a Model ID that I select from the live list, then generate one 1024 by 576 concept image for . Expected output: one Z-Image concept image and its caption. Verify that TEST CONNECTION succeeds before generating. ``` ### Caveat The README says the current Z-Image workflow is fixed to the qwen_3_4b and ae files, so a different diffusion checkpoint must be compatible with that text encoder and VAE or it will fail during sampling or decoding. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/karuvanan-minimax-h3-director-cut-studio/generate-design-concept-images-with-lm-studio/ ## How to split a song into vocal and music tracks locally with Mini Max H3 Director Cut Studio Repo: [karuvanan/MiniMax-H3-Director-Cut-Studio](https://github.com/karuvanan/MiniMax-H3-Director-Cut-Studio) | Difficulty: Intermediate | Time: 1 hour | Basis: From the README Run the Kim_Vocal_2 audio separator on a CUDA GPU to output Vocal, Music and original Mix files. **Who it is for:** A music creator or video editor with an NVIDIA CUDA GPU who needs clean vocal and backing tracks for a music video or dub. **The problem:** Isolating vocals from a mixed song usually means uploading files to a web service or accepting a lower-quality CPU separation. **The outcome:** Separate Vocal, Music and original Mix files produced by a local CUDA server, which the studio can use as audio inputs. ### Steps 1. Make sure the machine has a CUDA GPU and the complete `models/audio-separator` model and runtime folder. 2. Install the CUDA runtime with `.\install_audio_separator_cuda_runtime.bat`. 3. Start the local server with `.\start_audio_separator_server.bat`. 4. Open the studio with `.\run_h3_prompt_studio.bat` and use the AUDIO SEPARATOR button on the home page. 5. Check that the output Vocal and Music files have the expected duration and channels before using them. ### Prompt to paste into your agent ```text Using the repo karuvanan/MiniMax-H3-Director-Cut-Studio, install the CUDA runtime and start the Audio Separator server, then split