# No Hype: P-H3 Image, P-JiT, NitroFusion2, RelayVSR and Mirage

> Pruna cut the H3 image model to a 20 GB download that runs in 21 GB of VRAM; plus a 21,000-dollar image model and a 22 MB phone vision model.

- Published: 2026-11-12
- Author: Jacobo González Jaspe
- Language: en
- Tags: news, opinion, no-hype, ai
- Canonical URL: https://vorluxai.com/en/blog/sin-humo-2026-11-12-en/
- Source: VORLUX AI (J4SGON S.L.), https://vorluxai.com/

---

Pruna cut the H3 image model from a 143 GB download to 20 GB, and its int8 and 4-bit files run in ComfyUI in 21 GB of VRAM. This week is about image and video models that fit real machines, down to a 22 MB vision model for phones.

## Pruna shrinks the H3 image model to a 20 GB download

![Illustration generated with AI on our own machine.](https://vorluxai.com/assets/blog/sin-humo/2026-11-12-1.webp)
*Illustration generated with AI on our own machine.*


**What happened.** Pruna released P-H3-Image, a text-to-image and editing model based on MiniMax-H3 ([Hugging Face](https://huggingface.co/PrunaAI/P-H3-Image)). Pruna says its smallest quantized set is a 20 GB download instead of 143 GB for H3, the backbone went from 33B to 20B parameters, and the steps went from 50 to 8. On one B200 it takes 2.6 s instead of 12.4 s for a 4 MP image ([Pruna](https://www.pruna.ai/blog/oss-p-h3-image-launch)).

**What we think.** We think this is the release in this issue that matters most for a small studio. The model card says the int8 and 4-bit files run in ComfyUI in 21 GB of VRAM, while the full bf16 pipeline at 8 MP wants about 101 GB ([Hugging Face](https://huggingface.co/PrunaAI/P-H3-Image)). The 8-step model matches the 50-step original "for most prompts", in Pruna's words, so test the prompts that are yours.

**The good bit.** It edits with several input images, and the default text-to-image size is 2912 × 2912.

**What you can do.** If you have a 24 GB card, the quantized files are the ones to try. Our [ComfyUI local image stack guide](https://vorluxai.com/en/blog/comfyui-2026-local-image-generation-stack-guide-en/) covers the setup.

## Linum trains a 2B image model, P-JiT, for 21,000 dollars

![Illustration generated with AI on our own machine.](https://vorluxai.com/assets/blog/sin-humo/2026-11-12-2.webp)
*Illustration generated with AI on our own machine.*


**What happened.** Linum, a 2-person research lab, says its training run for P-JiT cost $21K ([Linum](https://www.linum.ai/field-notes/pjit-21k)). P-JiT is a 2B-parameter text-to-image model with a 4B text encoder taken from a frozen Qwen3.5-4B. Linum says the architecture converges in 11 times fewer samples than its own Linum v2.

**What we think.** We like a lab that publishes its training notes and its bill. A community ComfyUI conversion runs on ComfyUI's own sampler from 4 GB of free VRAM ([Hugging Face](https://huggingface.co/PedroMarinhoDev/P-JiT-21K-ComfyUI)). That conversion is not Linum's own release, so treat its numbers as one careful user's numbers.

**The good bit.** A 2B model in 4 GB is something an office laptop with a small graphics card can try.

**What you can do.** Use Linum's own settings to start: 50 steps, euler, simple, cfg 15 ([Hugging Face](https://huggingface.co/PedroMarinhoDev/P-JiT-21K-ComfyUI)). Then see which [hardware fits which job](https://vorluxai.com/en/products/).

## NitroFusion2 makes images in one or two steps

![Illustration generated with AI on our own machine.](https://vorluxai.com/assets/blog/sin-humo/2026-11-12-3.webp)
*Illustration generated with AI on our own machine.*


**What happened.** Dar-Yen Chen and Yi-Zhe Song released NitroFusion2, a 1 to 2 step text-to-image model built on Z-Image-Turbo ([GitHub](https://github.com/ChenDarYen/NitroFusion2)). The authors say it cuts distillation from the usual 4 to 8 steps to 1 or 2, and that at one step it renders sharp text and fine detail ([NitroFusion2](https://chendaryen.github.io/NitroFusion2.github.io/)). GGUF files range from 2.0 GB to 7.2 GB.

**What we think.** We read "one step" as a claim to check, not a feature to trust, and the authors say only that two steps "remains competitive". A community ComfyUI build reports about 1.7 times faster per step than bf16 on an RTX 4090, at half to a quarter of the size ([Hugging Face](https://huggingface.co/PedroMarinhoDev/NitroFusion2-ComfyUI)).

**The good bit.** Small GGUF files mean an older card can join in.

**What you can do.** Compare one step, two steps and your current model on ten of your own prompts before switching anything. Our [ComfyUI local image stack guide](https://vorluxai.com/en/blog/comfyui-2026-local-image-generation-stack-guide-en/) shows where the GGUF files go.

## RelayVSR doubles video resolution at about 29 frames a second on one GPU

![Illustration generated with AI on our own machine.](https://vorluxai.com/assets/blog/sin-humo/2026-11-12-4.webp)
*Illustration generated with AI on our own machine.*


**What happened.** Researchers at the University of Science and Technology of China released RelayVSR, which doubles a video's width and height ([RelayVSR](https://kopperx.github.io/RelayVSR/)). The authors measured 29.29 FPS at 1080p on one A100 80GB, with 13.82 GB peak GPU memory, against 7.80 FPS for FlashVSR-Tiny on the same card. A LightX2V build processed 243 frames in 5.348 seconds on one H100 80GB ([Hugging Face](https://huggingface.co/lightx2v/RelayVSR_X2)).

**What we think.** We think upscaling is where generated video gets cheaper first: make it small, then make it big. The model card says the design suppresses temporal flicker, the shimmer that gives away cheap upscaling ([Hugging Face](https://huggingface.co/lightx2v/RelayVSR_X2)). Every figure here comes from data-centre cards, not from a desktop.

**The good bit.** 13.82 GB of peak memory is within reach of workstation cards.

**What you can do.** Keep your clips short and upscale at the end. Check the memory against your card in our [VRAM calculator](https://vorluxai.com/en/tools/vram-explorer/).

## Cactus releases Mirage, a 22 MB vision model for phones

![Illustration generated with AI on our own machine.](https://vorluxai.com/assets/blog/sin-humo/2026-11-12-5.webp)
*Illustration generated with AI on our own machine.*


**What happened.** Cactus released Mirage, a small vision model with 51.3 million parameters that fits in about 22 MB using 2-bit quantization ([Cactus](https://cactuscompute.com/blog/mirage)). Cactus says it takes 74 ms end to end on a Samsung Galaxy S26 NPU. Mirage describes images and extracts structured data, and the same package includes Whistle, which transcribes speech in seven languages ([GitHub](https://github.com/cactus-compute/cactus)).

**What we think.** We like vision that runs where the picture is taken: a photo of a delivery note that never leaves the phone has no privacy policy to read. At 22 MB, this is a tool to read labels and forms, not a model that understands a whole contract.

**The good bit.** The cactus-compute package caches the engine and weights on first use and needs no JAX for local inference ([GitHub](https://github.com/cactus-compute/cactus)).

**What you can do.** If your staff photograph paperwork in the field, test it on twenty real photos. For bigger documents, see our [local AI for developers](https://vorluxai.com/en/developers/) page.

## In one line
P-H3-Image at 20 GB is the image model to try first on a 24 GB card.

## Work with us

If something in this column belongs in your business, on models that run on your own hardware rather than someone else's cloud, [tell us what you need](https://vorluxai.com/en/contact/) or see how we work in [consulting](https://vorluxai.com/en/consulting/).
