# Image Text to Video Models: Open-Weight AI Source: https://savrn.com/models/tasks/image-text-to-video Markdown alternate of the page above; the site index is https://savrn.com/llms.txt --- SAVRN Model Hub · Models by Task # Image Text to Video Models 8 open-weight image text to video models in the SAVRN Model Hub, with Jay, Unsloth AI and Rebel AI publishing the most. 8Models 7Publishers 33.1B to 33.1BParameter range 1Licenses ## SAVRN's Take One base model sits under all six entries. MiniMax-H3 takes text, images, video and audio as input and returns video with its own stereo soundtrack, up to 2K resolution and 15 seconds a clip, at 24 frames per second with 32 kHz audio. The other five repackage it: a GGUF quantization from Unsloth AI, two community builds from Jay in INT4, INT8, mixed and NVFP4 formats plus a pruned GGUF set, a LoRA adapter from Fal tuned for realistic people, and a hybrid checkpoint mirrored by Bruno Po covering text-to-video, reference-to-video and first-and-last-frame-to-video in ComfyUI. Downloads track that order: 4,449,605 a month for the base, 865,246 for the Unsloth GGUF, 209,828 and 143,202 for Jay's builds, 61,328 for Fal, 50,401 for the hybrid. The one parameter count in the set is 33.1B: 79.5 GB of memory at 16-bit, 39.7 GB at 8-bit, 19.9 GB at 4-bit. On the SAVRN Index the cheapest host at all three precisions is a single MI300X at $1.85 an hour, so cutting precision does not cut the bill; what it buys is a fit on the 16 GB to 24 GB consumer cards Jay's collection targets. The license reads other on all six, with no standard permissive grant to lean on. The hybrid's page says it is not an official MiniMax release and points to the upstream MiniMax license; every derivative carries the same tag, so read that document before production. Then pick the 79.5 GB base, a 19.9 GB 4-bit build, the 4-step or 8-step hybrid or the people LoRA by the job. SAVRN Research, 2026-09-18 ## Most Downloaded | Model | Publisher | Parameters | License | Monthly downloads | Cheapest GPUs at 16-bit | | --- | --- | --- | --- | --- | --- | | [MiniMax-H3](https://savrn.com/models/minimax-h3-2) | MiniMax | 33.1B | other | 5M | 1x MI300X, $1.85/hr | | [MiniMax-H3-GGUF](https://savrn.com/models/minimax-h3-gguf-2) | Unsloth AI | — | other | 2.1M | — | | [Minimax-H3-nvfp4-INT4-INT8-Convrot](https://savrn.com/models/minimax-h3-nvfp4-int4-int8-convrot) | Jay | — | other | 209.8k | — | | [MiniMax-H3-Pruned-GGUF](https://savrn.com/models/minimax-h3-pruned-gguf) | Jay | — | other | 143.2k | — | | [MiniMax-H3-Realism-People-LoRA](https://savrn.com/models/minimax-h3-realism-people-lora) | Fal | — | other | 61.3k | — | | [DaSiWa-MiniMax-H3-Hybrid](https://savrn.com/models/dasiwa-minimax-h3-hybrid) | Bruno Po | — | other | 50.4k | — | | [16bit-pixel-lora-minimax-h3](https://savrn.com/models/16bit-pixel-lora-minimax-h3) | Jin | — | other | — | — | | [FastH3-Omni_GGUFs](https://savrn.com/models/fasth3-omni-ggufs) | Rebel AI | — | other | — | — | ## Licenses | License | Models | Commercial use | | --- | --- | --- | | other | 8 | Read the license | ## Who Publishes Them | Publisher | Models | | --- | --- | | [Jay](https://savrn.com/model-publishers/abiray) | 2 | | [Unsloth AI](https://savrn.com/model-publishers/unsloth) | 1 | | [Rebel AI](https://savrn.com/model-publishers/realrebelai) | 1 | | [MiniMax](https://savrn.com/model-publishers/minimaxai) | 1 | | [Jin](https://savrn.com/model-publishers/leon1000) | 1 | | [Fal](https://savrn.com/model-publishers/fal) | 1 | ## All 8 Models Model · Image text to video ### [MiniMax-H3](https://savrn.com/models/minimax-h3-2) [MiniMax](https://savrn.com/model-publishers/minimaxai) Offical skills to improve prompt writing: skills on github Use MiniMax\-H3 directly via API\. Use MiniMax\-H3 directly via App\. MiniMax H3 is a general-purpose, omni-modal generative system. It supports unified understanding of multimodal contexts composed of text, images, video, and audio, and can generate video with native stereo audio at resolutions up to 2K and durations of up to 15 seconds. Thanks to its task-generalization-oriented system design, H3 already possesses broad multimodal context understanding and generation capabilities at the pre-training stage, enabling outstanding performance in following complex multimodal instructions. H3 supports the following input and output… Open weights other 33.1B parameters minimax-h3 [View model](https://savrn.com/models/minimax-h3-2) Model · Image text to video ### [MiniMax-H3-GGUF](https://savrn.com/models/minimax-h3-gguf-2) [Unsloth AI](https://savrn.com/model-publishers/unsloth) Instructions further below. GGUF for MiniMax-H3, compatible on most platforms including stablediffusion.cpp and Unsloth. You can run MiniMax-H3 via Unsloth: https://github.com/unslothai/unsloth/ GGUF quantizations of MiniMaxAI/MiniMax-H3 MiniMax H3 is an omni-modal generative system that produces video with native stereo audio, up to 15 seconds at 24 FPS with 32 kHz stereo audio. Both halves of the runtime are in this repo: the denoisers and the Qwen3-VL text encoder they need. H3 ships two denoisers, and which one you load decides what the model can be given: - fl2vapruned, the H3-Base first-and-last-frame variant. Text, plus zero, one or two frames. - ref2vapruned, the reference variant.… Open weights other gguf [View model](https://savrn.com/models/minimax-h3-gguf-2) Model · Image text to video ### [Minimax-H3-nvfp4-INT4-INT8-Convrot](https://savrn.com/models/minimax-h3-nvfp4-int4-int8-convrot) [Jay](https://savrn.com/model-publishers/abiray) This repository is a community-compiled collection of quantized and pruned weights for MiniMax H3 (Hailuo 3.0), optimized for local inference environments like ComfyUI. By unifying various quantization formats (INT4, INT8, Mixed, and NVFP4) into a single structured repository, this hub makes it easier for users with consumer GPUs (16GB - 24GB VRAM) to experiment with MiniMax H3's powerful omni-modal text/image/audio-to-video generation capabilities. If you are new to local generation and aren't sure what to download, use this guide based on your graphics card. Perfect for RTX 4070 Ti Super, RTX 4080, etc. Perfect for RTX 3090, RTX 4090, etc. Exclusively for RTX 5090, PRO 6000, and other… Open weights other diffusers [View model](https://savrn.com/models/minimax-h3-nvfp4-int4-int8-convrot) Model · Image text to video ### [MiniMax-H3-Pruned-GGUF](https://savrn.com/models/minimax-h3-pruned-gguf) [Jay](https://savrn.com/model-publishers/abiray) This repository (Abiray/MiniMax-H3-Pruned-GGUF) provides pruned and quantized GGUF weights for the MiniMax H3 omni-modal generative model. MiniMax H3 is designed for unified multimodal context processing, capable of generating synchronized high-definition video and 32 kHz stereo audio from text, image, audio, and video inputs. 1. Download your desired.gguf variant from the table above. 2. Place the downloaded.gguf file into the ComfyUI/models/unet/ directory. 3. In your ComfyUI workflow, load the model using the UnetLoaderGGUF node. MiniMax H3 is released under the MiniMax H3 Community License Agreement. Please refer to the MiniMaxAI/MiniMax-H3 repository and the repository's LICENSE file… Open weights other [View model](https://savrn.com/models/minimax-h3-pruned-gguf) Model · Image text to video ### [MiniMax-H3-Realism-People-LoRA](https://savrn.com/models/minimax-h3-realism-people-lora) [Fal](https://savrn.com/model-publishers/fal) A LoRA adapter for MiniMax H3 specialized in realistic people: faces that hold up in close-up, natural skin texture, believable expressions and gestures, film-style lighting and documentary camera movement. Same prompt, same seed — base model on the left, this adapter on the right: 19 pairs, same prompt, same seed, adapter on vs off — the only variable is the LoRA. The trigger word is present on both sides, so it is not doing the work. Each pair plays the base model first, then freezes and dims while the adapted version plays beside it. Close-up talking faces, arguments, several people speaking at once, weathered skin, children, ritual and travel scenes. Nothing cherry-picked from a larger… Open weights other minimax-h3 [View model](https://savrn.com/models/minimax-h3-realism-people-lora) Model · Image text to video ### [DaSiWa-MiniMax-H3-Hybrid](https://savrn.com/models/dasiwa-minimax-h3-hybrid) [Bruno Po](https://savrn.com/model-publishers/brurpo) Unmodified 4-step and 8-step DaSiWa MiniMax H3 Hybrid SafeTensors checkpoints mirrored for BRP Canvas downloads. The hybrid checkpoint supports text-to-video, reference-to-video, and first/last-frame-to-video through the same ComfyUI workflow. This is not an official MiniMax or DaSiWa distribution. Review the original model page and the included upstream MiniMax license before use. BRP Canvas defaults to shift video 9 and shift audio 4 for both distilled checkpoints. - 4-step: 56c52c7890c105308d28fe9c25c25fdb80e6cd6a54e2604d8af71732ba4ed74f - 8-step: e0441d26414f6e0c28f43d580e6cc56fad424da0fa4d261b698ca73188aa6332 Open weights other minimax-h3 [View model](https://savrn.com/models/dasiwa-minimax-h3-hybrid) Model · Image text to video ### [16bit-pixel-lora-minimax-h3](https://savrn.com/models/16bit-pixel-lora-minimax-h3) [Jin](https://savrn.com/model-publishers/leon1000) A style LoRA that renders MiniMax H3 footage as 16-bit pixel-art video — large square pixel clusters, staircase outlines, a limited palette, flat two-tone shading and distinct held sprite poses, like a SNES-era cutscene in motion. H3 trains audio jointly, so sound comes through with the picture. This repo hosts the weights for the fal endpoint, which is the supported way to use this style — the endpoint applies the tuned trigger phrasing and Raw single take, generated in one pass — no cherry-picking. Adapter weights for MiniMaxAI/MiniMax-H3, distributed under the Open weights other minimax-h3 [View model](https://savrn.com/models/16bit-pixel-lora-minimax-h3) Model · Image text to video ### [FastH3-Omni_GGUFs](https://savrn.com/models/fasth3-omni-ggufs) [Rebel AI](https://savrn.com/model-publishers/realrebelai) Open weights other comfyui [View model](https://savrn.com/models/fasth3-omni-ggufs) ## Questions ### Which Image text to video models are most downloaded? By monthly downloads reported by the Hugging Face Hub: MiniMax-H3 (5M); MiniMax-H3-GGUF (2.1M); Minimax-H3-nvfp4-INT4-INT8-Convrot (209.8k). ## Other Tasks - [Text Generation](https://savrn.com/models/tasks/text-generation) 1506 - [Image and Text to Text](https://savrn.com/models/tasks/image-and-text-to-text) 560 - [Feature Extraction](https://savrn.com/models/tasks/feature-extraction) 396 - [Robotics](https://savrn.com/models/tasks/robotics) 379 - [Text Classification](https://savrn.com/models/tasks/text-classification) 328 - [Text to Image](https://savrn.com/models/tasks/text-to-image) 228 - [Speech Recognition](https://savrn.com/models/tasks/speech-recognition) 185 - [Text to Speech](https://savrn.com/models/tasks/text-to-speech) 143 - [Image Classification](https://savrn.com/models/tasks/image-classification) 139 - [Audio Classification](https://savrn.com/models/tasks/audio-classification) 130 - [Sentence Similarity](https://savrn.com/models/tasks/sentence-similarity) 103 - [Token Classification](https://savrn.com/models/tasks/token-classification) 99 [See all](https://savrn.com/models/tasks)