logo
RunComfy
  • ComfyUI
  • Trainer
  • Models
  • MCPNew
  • Pricing
  • Seedance 2.5Hot
  • Seedream 5.5New

Workflows by Model

HotMiniMax H3Video with audio from text, images or referencesHotQwen Image 2.1Create images and edit photos with referencesWanVideo from text or images, motion transferFLUXImage generation, editing and character variationsLTXVideo with synced audio from text or imagesQwen Image EditPhoto editing with text and image referencesKrea 2Image creation, style transfer and reference editsZ-ImageFast portraits and concepts from text

Popular & New Workflows

NewMiniMax H3 Character Swap V1.5Swap video characters, keep the original audioNewP-H3-Image Reference EditEdit a photo guided by a reference imageHotMiniMax H3 Image-to-VideoOne image to cinematic video with synced audioHotLTX 2.5 Image-to-VideoOne frame to sharp video with synced audioHotMiniMax H3 Character Swap (SAM3)Swap video characters while following the motionQwen Image 2.1 Multi-Image EditingPrecise scenes from up to 10 reference imagesFLUX Klein Face SwapNatural, lifelike face swapsSeedVR2 Video UpscalerSharp, smooth video upscaling
View More Workflows

Train a Model

NewMOSS-Image BaseImage generation, newly supportedNewFLUX.3 KleinImage generation, newly supportedHotKrea 2The most-trained image modelHotKrea 2 TurboFew-step image generationHotQwen Image 2.1Image generation and editingZ-Image TurboFast portraits and concept imagesMiniMax H3Video with synchronized audio

Train & Deploy

Getting StartedYour first training job with Ostris AI ToolkitDatasetsUpload and manage training datasetsInferenceRun your trained models in a playground or via APIDeploymentsServe your trained models as dedicated endpoints
Train More Models

Image Models

NewGPT Image 2.65% OFF · Detailed text to image, GIF and SVG, up to 4KNewSeedream 5.510% OFF · Text to image up to 2K, with transparent PNGsHotSeedream 5.0 Pro41% OFF · Reference edits with multilingual textHotSeedream 4.520% OFF · Photoreal edits with brand consistencyHotWan 2.6 ImageInstruction-based image editingFLUX.2Precise, photorealistic image editsNano Banana Pro77% OFF · Precise 2K to 4K visuals from sketchesQwen Image EditLoRA-based, structure-aware image editing

Video Models

NewSeedance 2.5 Flash34% OFF · Text, image, or reference to videoNewKling 4.010% OFF · Animate a still into a 3 to 30s clipHotWan 3.0 Prime13% OFF · Fast clips from reference videosHotSeedance 2.536% OFF · Longer, more consistent cinematic videoHotWan 3.0Clips from image, video and audio referencesHotMiniMax H3Animate a first frame into a 5 to 15s clipWan 2.510% OFF · Fluid motion with native audioFLUX 3 MaxClips from text, a still, or keyframes
View More Models

API

Serverless (ComfyUI)Deploy ComfyUI workflows as serverless endpoints.Serverless (LoRA)Deploy LoRAs as serverless endpoints.Model APIRun hosted models on-demand with no deployment.Trainer APITrain, manage, and retrieve LoRAs via API.CLIRun RunComfy models from your terminal.
MODELS
Explore
All Models
LIBRARY
Generations
MODEL APIS
API Docs
API Keys
ACCOUNT
Usage

MiniMax H3 Max Reference to video: Multimodal Video on Models and API | RunComfy

minimax/minimax-h3-max/reference-to-video

MiniMax H3 Max Reference to video turns image, video, and audio references plus a prompt into 5-15s 480p/768p clips with native audio.

Describe the scene, motion, camera, and audio. Refer to assets as Image 1, Image 2, Video 1, Audio 1, and say what each reference supplies.
Image 1
Image 2
Subject or style reference image URLs, cited in the prompt as Image 1, Image 2, and so on. Combined with videos and audios, at most 12 reference files.
Motion reference clips (about 2-15s each; combined duration at most 15s), cited as Video 1, Video 2. Combined with images and audios, at most 12 reference files.
Optional voice or ambience clips (about 2-15s each; combined duration at most 15s). Cannot be the only reference; provide at least one image or video with them.
Output aspect ratio. adaptive follows the reference framing when possible.
Native generation resolution. 480p is faster/cheaper; 768p is the sharper default.
Length of the generated video in seconds (5-15).
How much effort to spend rewriting the prompt before generation. balanced returns quickly; quality spends longer on a richer prompt.
Fixed seed for reproducible results. Use -1 for a random seed.
If set to true, the safety checker will be enabled.
Idle
Output video is billed at $0.09 per second, plus $0.20 per second of reference video (rounded up to the next whole second), plus $0.03 per reference image. Reference audio is free. Example: a 5s clip with a 5s reference video and 2 reference images costs $1.51.

Introduction To MiniMax H3 Max Reference to video

MiniMax's MiniMax H3 Max Reference to video builds a 480p or 768p clip from a written brief plus the image, video, and audio references you attach, rendering picture and native audio in one pass.

Trading lookalike casting and manual motion matching for references the model reads directly, MiniMax H3 Max Reference to video helps brand teams, agencies, and previz artists hold a character, product, or camera language steady across shots.

For developers, MiniMax H3 Max Reference to video on RunComfy can be used both in the browser and via an HTTP API, so you don't need to host or scale the model yourself.

Ideal for: Character Consistency | Product Reveals | Style Matching

MiniMax / MiniMax H3 Max Reference To Video#


Parameters#


ParameterRequiredTypeDefaultRange / OptionsDescription
prompt *Yes (*)stringSample briefTextScene, motion, camera, and audio. Cite Image 1 / Video 1 / Audio 1 and say what each reference supplies.
reference_images *Yes (*)array (image URL)Sample stillUp to 12 combined refsSubject or style images cited as Image 1, Image 2, …
reference_videosNoarray (video URL)NoneUp to 12 combined refs; ~2-15s each, ≤15s combinedMotion clips cited as Video 1, Video 2, …
reference_audiosNoarray (audio URL)NoneUp to 12 combined refs; ~2-15s each, ≤15s combinedVoice or ambience; cannot be the only reference.
aspect_ratioNostringadaptiveadaptive, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16Output framing.
resolutionNostring768p480p, 768pNative generation resolution.
durationNointeger55-15Clip length in whole seconds.
prompt_expansion_mode *Yes (*)stringbalancedbalanced, qualityPrompt rewrite effort before generation.
seedNointeger-1 (random)IntegerFixed seed for reproducibility.
enable_safety_checkerNobooleantruetrue, falseEnables the safety checker when true.

Pricing#


ItemPrice
Output video$0.09 per second
Reference video$0.20 per second (rounded up to the next whole second)
Reference image$0.03 per image
Reference audioFree

Resolution does not change the per-second rate. Example: a 5s clip with a 5s reference video and 2 reference images costs $1.51. Check the Generation section on this page for the live credit estimate.

Related Models

ltx-2.5/audio-to-video/fast

LTX 2.5 Fast audio-to-video for track-timed 1080p clips

ltx-2/fast/text-to-video

Next-gen tool turning prompts into cinematic 1080p video clips with audio

wan-2.7/edit-video

AI-driven footage transformation with stable motion and design control

seedance-1.0/text-to-video

Generate cinematic videos from text prompts with Seedance 1.0.

wan-3.0-prime/image-to-video

Wan 3.0 Prime animates a first-frame image into video fast

pixverse-v6/image-to-video

Create controlled 1080p clips from images, text, references, or video.

Frequently Asked Questions

What is MiniMax H3 Max Reference to video used for?

MiniMax H3 Max Reference to video generates short clips from a prompt plus image, video, and/or audio references. It is a strong fit when you need character, product, or style consistency across shots while picture and native audio are produced together.

How does MiniMax H3 Max Reference to video differ from plain image-to-video?

Image-to-video usually animates one opening frame. MiniMax H3 Max Reference to video conditions on multiple multimodal references—images for identity or look, optional videos for motion, optional audio for voice or ambience—so you can steer several cues in one brief.

What input limits should I know before using MiniMax H3 Max Reference to video?

Clips are 5–15 seconds at 480p or 768p. Reference images, videos, and audios together may total at most 12 files; video and audio refs are typically about 2–15 seconds each with combined duration up to 15 seconds. Audio cannot be the only reference—include at least one image or video. Check the parameter panel for live limits.

How should I write prompts for MiniMax H3 Max Reference to video?

Name each asset in order (Image 1, Video 1, Audio 1) and say what it supplies—identity, wardrobe, motion, or voice. Add clear camera and audio lines so MiniMax H3 Max Reference to video keeps lip-sync, ambience, and framing aligned with the references.

Does MiniMax H3 Max Reference to video generate audio with the picture?

Yes. MiniMax H3 Max Reference to video renders native audio in the same pass as the video. You can also attach reference audio when you need a specific voice or ambience, as long as an image or video reference is present too.

What resolutions and aspect ratios does MiniMax H3 Max Reference to video support?

MiniMax H3 Max Reference to video supports 480p and 768p, with aspect ratios including adaptive, 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16. Adaptive follows the reference framing when possible.

Can developers use MiniMax H3 Max Reference to video through the RunComfy API?

Yes. Prototype MiniMax H3 Max Reference to video in the RunComfy Web UI, then call the same model and parameters over the RunComfy HTTP API for automation. New accounts typically receive a free trial USD balance to start testing.

How much does it cost to generate with MiniMax H3 Max Reference to video on RunComfy?

On RunComfy, MiniMax H3 Max Reference to video is billed at $0.09 per second of output video, plus $0.20 per second of reference video (rounded up to the next whole second), plus $0.03 per reference image. Reference audio is free. For example, a 5-second clip with a 5-second reference video and 2 reference images costs $1.51. Generations consume USD/credits; check the Generation section on the model page for the live estimate.

Follow us
  • LinkedIn
  • Facebook
  • Instagram
  • Twitter
Support
  • Discord
  • Email
  • System Status
  • Affiliate
Video Models
  • Seedance 2.5 Flash Text To Video
  • Seedance 2.5 Flash
  • Seedance 2.5 Flash Reference to Video
  • Wan 3.0
  • MiniMax H3 Fast
  • Wan 2.5
  • View All Models →
Image Models
  • Seedream 5.5
  • Seedream 5.5 Edit
  • Seedream 5.5 Layer Decomposition
  • Wan 2.6 Image to Image
  • Nano Banana Pro
  • Seedream 5.0 Pro
  • View All Models →
Legal
  • Terms of Service
  • Privacy Policy
  • Cookie Policy
RunComfy
Copyright 2026 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.

Examples Of MiniMax H3 Max Reference to video

Video thumbnail
Loading...
Video thumbnail
Loading...
Video thumbnail
Loading...
Video thumbnail
Loading...
Video thumbnail
Loading...
Video thumbnail
Loading...