Launch Mix Studio
Run the installer on Windows or macOS. On Linux, clone the repository and run ./start-mixstudio.sh. Mix Studio opens setup automatically.
Your studio.
On your machine.
A clean, responsive AI workspace built on ComfyUI. Run highly tuned image and video workflows flawlessly from your desktop or your phone. Features curated setups for Krea 2, Flux Klein, Qwen Edit, MiniMax H3, LTX 2.3, Wan 2.2, and SCAIL 2.
ComfyUI
Tailscale
The right models, already tunedCurated model families and workflow-tested defaults are chosen to produce strong results with less trial and error.
Create from anywhereUse the same studio from your phone at home, on the couch, or away from your desk.
Designed for every screenPrompt, edit, queue, review, and reuse work from the full desktop workspace or a phone-sized interface.
Created by Black Mixture
A creative studio for artists working with AI.
Mix Studio is built by Black Mixture: a creative studio and education platform focused on design, storytelling, and practical AI tools for artists and innovators.
A different local AI workflow
The computer does the work. Your phone runs the studio.
Most local tools chain you to a desk. Mix Studio lets your PC handle the heavy lifting while giving you full creative control from anywhere.
GPU power stays homeComfyUI, models, storage, and every generation remain on the generation computer.
Your studio stays within reachTailscale connects the same studio without moving the workload.
Create anywherePrompt, queue, review, and reuse work from the same responsive UI on desktop or phone.
Quick start
Open the app. Finish setup in one place.
Connect ComfyUI
Start a detected installation, install ComfyUI Desktop, or choose a live endpoint. Then use Quick setup, install only that workflow, or open the full guide.
Carry it to your phone
Add an Owner PIN, then use the Phone access card to install Tailscale and copy or share the private URL.
Curated for better results
Leading models. Optimized settings.
Mix Studio starts with carefully selected models and practical, workflow-tested defaults. You get strong results quickly, while advanced controls remain available when you need them.
Create images
Generate with Krea 2, optimized sampling, source matching, depth guidance, LoRAs, and prompt assistance.
Edit and outpaint
Address references directly in the prompt with @, use masks and smart selection, choose Continue Edit, or expand a source while preserving its detail.
Ocean
Island
Snow biome
Lava volcano
Direct with regions
Position the ocean, island, snow biome, and lava volcano independently, then generate one coherent scene.
Generate video
Use MiniMax H3 for text, first/last-frame, or reference-driven video with native audio, alongside LTX Face ID and lipsync audio, Wan, 10Eros, and SCAIL.
Finish and compare
Upscale with tuned SeedVR2 or Ultimate SD profiles, then reveal, zoom, and pan across details before exporting.
Keep a usable library
Search, sort, like, group, continue editing, document, and mirror exports from the mobile or desktop library.
Measured, compared, tuned
Benchmarked by workflow. Tuned for the result.
There is no single best local model. Mix Studio evaluates each family against the job it needs to do, then balances visual quality, iteration time, control, stability, and hardware fit.
Top model by jobCurrent tuned lineup
- #1Image CreationFast DefaultKrea 2 Turbo
12 Steps · Source and Depth Guidance · Regional Control
- #1Precision EditingInstruction FitFlux 2 Klein 9B
4-Step Scheduler · Up to 3 References · Masks and Outpaint
- #1Cinematic VideoAV PipelineLTX 2.3
Two-Stage Refinement · Image or Text · Native Audio
- NewOmni VideoNative ComfyUIMiniMax H3
Text + Frames · Optional References · Native Stereo Audio
- #1Motion TransferTracking FitSCAIL 2
6-Step Workflow · SAM3 Tracking · Infinity Long Clips
VRAM guidanceOffload tier · Recommended
Mix Studio does not enforce a VRAM cutoff or system RAM requirement. Klein 4B begins at 4 GB with offload. Krea 2 and most video workflows begin at an 8 GB guided tier; MiniMax H3 begins at 12 GB. Quantization, resolution, duration, and offload all affect fit.
04 GB8 GB16 GB24 GB
Entry EditFlux 2 Klein 4B FP8 · RAM offload
Core ImageKrea 2 + SeedVR2
Precision EditFlux 2 Klein 9B
Cinematic VideoLTX 2.3 · aggressive offload
Omni VideoMiniMax H3 · NVIDIA
Motion TransferSCAIL 2 · FP8 or manual GGUF
RTX PRO 6000 Blackwell · 96 GB VRAM · Jul 2026Measured Timing
Warm runs. Resolution, duration, and offload affect speed.
Image≤ 9 secKrea 2
Edit≈ 5 secKlein 9B
Short Video≈ 1–2 minLTX / SCAIL
Long Video≈ 8 minLTX / SCAIL
Free & Open Source
Made in Mix Studio
Real outputs, straight from the queue.
Every example was generated locally through tuned workflows for precise image edits, motion-preserving video transformations, image animation, and box-directed compositions.
SCAIL 2 · directed video editing
Change the subject, object, or world
Transfer the original motion while preserving timing, camera movement, realistic physics, and scene-aware lighting.
Edit · reference-addressed prompt
Use @ to assign a role to each input
@Image 1 supplies the character, @Image 2 supplies the jacket, and @Image 3 supplies the setting in one Flux Klein 9B edit.
Sky
Cliff
Sea monster
Car
Region prompting · plan to result
Define the scene. Generate the composition.
Place the sky, cliff, sea monster, and car with visual regions, then generate the complete scene in one pass.
Depth guide
New image, same 3D structure
Source → Depth Anything V3 map → re-imagined scene. Every armchair and beam stays in place.
Inpaint
Paint the mask, change only the mask
A character LoRA swapped into the masked area while the wave doesn't move a drop.
SCAIL 2 · style transfer
From rendered world to wireframe motion
The subject and movement stay readable while the entire visual language changes.
Outpaint
Square to 21:9, seamlessly
The original center holds while the room expands naturally on both sides.
Useful when you need it
The details stay out of the way.
Profiles, contextual guidance, and dependency checks are close at hand without taking over the creative workspace.
Profiles
Who’s creating?
Optional PINs keep galleries, folders, presets, and history separate for each person.
Guidance
Help appears in context
Focused tips explain gestures and controls when they become useful.
Setup health
Know what is ready
Each workflow reports the models and nodes it needs, with setup actions in the same UI.
ComfyUIYour current installation
DetectedComfyUI installReady to connect
Downloaded modelsReady to reuse
Nothing needs to move.Choose folders manually
Existing setup
Use your ComfyUI and models
Mix Studio can auto-detect your current ComfyUI install and downloaded models through registered files, shared-model YAML, and Desktop 2 directories across Windows, macOS, and Linux.
What setup checks and installs
Hardware-aware setup detects GPU memory, offers the official Klein 4B FP8 route with offload from 4 GB, and offers Krea 2 INT8 ConvRot on 4–12 GB systems when Krea is selected. It does not enforce a system RAM requirement. It verifies ComfyUI 0.26.0 or newer for Krea 2, 0.27.0 or newer for native INT8 ConvRot, and 0.30.0 or newer for MiniMax H3. H3's separate Ref2VA model downloads only when Reference mode is selected. Setup reuses models already registered through ComfyUI, warns before a below-tier choice while keeping the install-anyway option, and never silently lowers a requested resolution or batch. The bootstrap retrieves the public Mix Studio repository; ComfyUI, models, and nodes download only after you choose an in-app setup option. The uninstaller leaves shared ComfyUI files and models in place.
Common questions
- Is Mix Studio free?
- Yes. Mix Studio is free to download. Model licenses still apply.
- Can I use my ComfyUI install?
- Yes. Auto-detect it or choose its folder manually.
- What works on Apple Silicon?
- Mix Studio supports a source-based ComfyUI environment on macOS, native folder selection, Metal-aware setup, Krea image workflows, and official BF16 LTX 2.3. MiniMax H3 and the curated 10Eros, Wan 2.2, and SCAIL 2 routes are disabled on Metal.
- What works on Linux and AMD?
- Linux with NVIDIA supports the full curated workflow set, including MiniMax H3, through an existing source ComfyUI or an explicitly configured user service. AMD ROCm on Windows or Linux is experimental and requires a working ROCm-enabled ComfyUI/PyTorch environment; Mix Studio applies AMD-safe model and attention defaults, but the curated H3 route remains NVIDIA-only.
- Can I reuse downloaded models?
- Yes. Registered models, manually selected folders, shared-model YAML, and Desktop 2 model directories are supported.
- Can I use a 4 GB GPU?
- Yes. Flux 2 Klein 4B FP8 is the 4 GB entry route when current ComfyUI can offload the model and encoder. It will be slower, Krea 2 still starts at 8 GB, and larger workflows may need stronger offloading or quantized weights.
- Can I generate video with 8 GB of VRAM?
- You can try most video routes. Mix Studio treats 8 GB as an experimental offload tier and keeps the install available without a system RAM requirement. MiniMax H3 starts at a 12 GB guided tier. Begin with short, smaller videos; the curated graphs recommend 24 GB VRAM.
- What devices can I use?
- Run it on Windows, Linux, or Apple Silicon macOS, then create from desktop, tablet, or phone. Metal supports a smaller workflow set, while AMD ROCm support is experimental.
- Does it work away from home?
- Yes. Tailscale keeps your studio privately within reach.
- Does my work stay local?
- Yes. Generations and your gallery remain on your machine.
Desktop power, mobile freedom