2026-11-11 · Qwen Image 2.1 · 1K · 8
MiniMax H3 Image — чекпоинт изображения. Эта форма — Qwen Image 2.1
MiniMax H3 Image — чекпоинт изображения на карточке MiniMax. Введение говорит примерно 16 megapixels. Раздел размеров говорит примерно от 4 MP до 20 MP. Эта форма — Qwen Image 2.1 в 1K.
Эта форма — Qwen Image 2.1 в 1K за 8 кредитов. Она не запускает веса MiniMax H3 Image, не ComfyUI и не локальный GPU. 64 кредита — это 8 изображений.
64 бесплатных кредита = 8 стандартных изображений. Карта не нужна.
Доска результатов
Здесь появятся сгенерированные варианты.
- Разработчик
- MiniMax
- Доступно на Nano AI
- Qwen Image 2.1 · 8 кредитов за изображение
- Страница обновлена
- 11 ноября 2026 г.
Источник: huggingface.co
Photographs, not MiniMax H3 Image outputs
David S. Soriano, CC BY-SA 4.0, 2023-05-29. The tube reads CRAFTER'S CLOSET and WATERCOLOR. No hands. Andreas Praefcke, public domain. The date field says 2011-06 and the file timestamp is 2011-06-29. A cropped red mark is not a readable word. Not Qwen Image 2.1 outputs.
Что печатает карточка MiniMax H3 Image и что отправляет эта кнопка
MiniMax H3 Image — это чекпоинт изображения MiniMax, карточка прочитана 2026-11-11. The model card, read 2026-11-11, says about 16 megapixels, from 4096x4096 to 6272x2688, without an upscaler, distilled to 6 steps with no classifier-free guidance. The sizes section on the same card says the working range is about 4 MP to 20 MP, best around 16 MP. min_generation_area defaults to 4194304, which the card writes as 2048x2048, and a smaller size is resized back with LANCZOS. The zoo line says 68.65 GB, and the optional PE file is 17.9 GB. The file table says the transformer is 66.28 GB and the VAE is 2.34 GB. The text encoder, loaded from MiniMax-H3, is 66.7 GB and is not inside the 68.65 GB line. The highlights say up to 16 reference images. This form is Qwen Image 2.1 at 1K for 8 credits. It is not those weights, not ComfyUI, and not a local GPU run. 64 credits are 8 images. /p-h3-image is Pruna's pack, with the 55 GB, 30 GB and 20 GB figures already on that page. Motion on this site is Hailuo 02 Standard, 6 seconds, 60 credits: text to video does not send a resolution, and image to video sends 768P.
Image to video on this site stays on /minimax-h3 with Hailuo 02 Standard. /minimax-h3-max is the fast video tier and uses the same Hailuo form. /minimax-h3-vbvr is a third-party LoRA note, still Hailuo 02 Standard. /hyperflow-h3-flash is Video Rebirth's distilled clip, not this still. None of those pages is this checkpoint.
MiniMax H3 Image next to this form
- Card, read 2026-11-11
- about 16 megapixels · 4096x4096 to 6272x2688 · 6 steps · no upscaler in that sentence
- Sizes section
- about 4 MP to 20 MP · best around 16 MP · LANCZOS · 4194304 · 2048x2048
- Files
- 68.65 GB · PE 17.9 GB · transformer 66.28 GB · VAE 2.34 GB · text encoder 66.7 GB
- References
- up to 16 reference images on the highlights line
- This form
- Qwen Image 2.1 · 1K · 8 credits · 64 credits are 8 images
- Not this page
- /p-h3-image is 55 GB, 30 GB and 20 GB · Hailuo 02 Standard is 6 seconds and 60 credits, image to video 768P
Keep the still, the pack, and the clip apart
The card's 16 MP and its 4 MP line
About 16 megapixels, 4096x4096 to 6272x2688, without an upscaler. The same card also says about 4 MP to 20 MP, and min_generation_area 4194304 (2048x2048) with LANCZOS.
68.65 GB is not the only size line
Zoo: 68.65 GB, optional PE 17.9 GB. File table: transformer 66.28 GB, VAE 2.34 GB, text encoder 66.7 GB from MiniMax-H3. Up to 16 reference images.
This button and the sibling pages
Qwen Image 2.1, 1K, 8 credits. /p-h3-image keeps 55 GB, 30 GB and 20 GB. Hailuo 02 Standard is 6 seconds and 60 credits, and image to video sends 768P.
MiniMax H3 Image, P-H3-Image, and this form
Read 2026-11-11. /p-h3-image stays the VRAM-pack guide. This page does not repeat that install.
| MiniMax H3 Image card | /p-h3-image | This form | |
|---|---|---|---|
| Job | text to image and image editing | Pruna's ComfyUI pack | Qwen Image 2.1 still |
| Size | about 16 megapixels, and also about 4 MP to 20 MP | 55 GB, 30 GB, 20 GB | 1K |
| Steps | 6, or the request is rejected | 8 on that card | not a step control |
| Price here | not hosted | the guide opens Qwen Image 2.1 | 8 credits |
| Motion | no audio layers in the DiT sentence | not a video page | Hailuo 02 Standard, 6 seconds, 60 credits, image to video 768P |
How to make a still on this page
- 1
Write a scene with no person
The lines below are original. They do not ask for a 4096x4096 file and they do not load a checkpoint.
- 2
Leave the model on Qwen Image 2.1
The picture is 1K and costs 8 credits. 64 credits are 8 images. This is not 16 megapixels and not 6 steps.
- 3
Open another page for a different job
ComfyUI sizes stay on /p-h3-image. A clip is Hailuo 02 Standard on /minimax-h3, 6 seconds, 60 credits.
Prompts for the 1K still
Original lines for Qwen Image 2.1. They are not the model card's showcase prompts.
Paint pans
A white tray of watercolor pans on a bare table, one brush beside it, no hands and no logos.
Green plate
A white plate with thin green arcs on a linen cloth, overhead light, no food and no text.
Clay cup
One unglazed clay cup on a stone sill, morning light, no person and no lettering.
Paper sheet
A blank sheet of paper on a wooden board, one pencil, no writing and no hands.
Вопросы и ответы
MiniMax H3 Image questions
The card's own lines stay side by side. This form's price is 8 credits.
Эта форма вызывает MiniMax H3 Image?
Нет. Карточка описывает те веса. Эта форма — Qwen Image 2.1 в 1K за 8 кредитов. Это не ComfyUI.
Why do 16 MP and 4 MP both stay?
The introduction says about 16 megapixels, from 4096x4096 to 6272x2688, without an upscaler. The sizes section says about 4 MP to 20 MP, and min_generation_area defaults to 4194304 (2048x2048) with LANCZOS. This page does not pick one.
Is 68.65 GB the whole download?
The zoo line says 68.65 GB, plus an optional PE of 17.9 GB. The file table says transformer 66.28 GB and VAE 2.34 GB. The text encoder is 66.7 GB and comes from MiniMax-H3. Those lines are not added together here.
Is this the same page as /p-h3-image?
No. /p-h3-image is Pruna's pack at 55 GB, 30 GB and 20 GB. This page is the MiniMax card. Both open Qwen Image 2.1 at 1K for 8 credits. Neither runs the weights.
Can this page do image to video?
The still form cannot. Hailuo 02 Standard on /minimax-h3 is 6 seconds and 60 credits. Text to video does not send a resolution. Image to video sends 768P. That is not 16 reference images and not synced audio.
Is Nano AI part of MiniMax?
No. Nano AI is not affiliated with MiniMax. The name only identifies the checkpoint.
Ещё на Nano AI
Related pages
Nano AI не связан с MiniMax. Имя только называет чекпоинт. Логотип MiniMax не показывается.