Skip to content
Vellria

MiniMax H3 Image Uncensored

Generates from a prompt at 1K in seven shapes, or edits from up to 9 references at 1K.

ImageText to image · Image to imageMiniMax6 credits
vellria-minimax-h3-image
Docs

MiniMax H3 Image is developed by MiniMax. Vellria provides access to the model; it does not own it. All trademarks belong to their respective owners.

Input

Form

0 / 10000

Reference

0–9 images. You can upload files or paste https URLs; the provider downloads the image from the URL.

Settings

Leave empty for a random value between 0 and 2147483647.

Sign in to run

No generation yet

No generation yet

Output

No generation yet

Cover art

6 credits

Sign in to run

Your request will cost 6 credits.

Overview

This family keeps the setup short: there is a single resolution tier, so the only decision about the output is its shape, picked from a fixed list that runs from square out to wide and tall frames. Every shape costs the same, which means the frame can follow the picture rather than the budget — choose the proportion the image will actually be used at and let the prompt carry the rest. Both modes take a seed, so when a draw is worth coming back to, note it along with the wording.

Prompt or reference

With only a description, the model draws the frame from nothing and the shape is yours to pick. Attach a picture and it edits instead, accepting several at once; every image past the first adds to what the run costs, so hand over what the edit needs rather than everything you have. Editing offers the same list of shapes, plus an adaptive setting it starts on, which follows the proportion of the first picture you hand over.

Specifications

  • Size (Text to image): 1024×1024 (1:1)1248×832 (3:2)832×1248 (2:3)1184×896 (4:3)896×1184 (3:4)1376×768 (16:9)768×1376 (9:16)
  • Size (Image to image): 1K · Matches input image1024×1024 (1:1)1248×832 (3:2)832×1248 (2:3)1184×896 (4:3)896×1184 (3:4)1376×768 (16:9)768×1376 (9:16)
  • Reference images: 1–9 reference images

How the image side of the catalog works as a whole, and what it allows, is on the uncensored AI image generator page; changing a picture you already have is covered by the uncensored image editor.

Pricing

Text to image

SizeCredits
1024×1024 (1:1)6 credits
1248×832 (3:2)6 credits
832×1248 (2:3)6 credits
1184×896 (4:3)6 credits
896×1184 (3:4)6 credits
1376×768 (16:9)6 credits
768×1376 (9:16)6 credits

Image to image

SizeCredits
1K · Matches input image6 credits
1024×1024 (1:1)6 credits
1248×832 (3:2)6 credits
832×1248 (2:3)6 credits
1184×896 (4:3)6 credits
896×1184 (3:4)6 credits
1376×768 (16:9)6 credits
768×1376 (9:16)6 credits

+1 credit per extra image

Call it from the API

Send vellria-minimax-h3-image as the model field to POST /v1/generate/image, with your API key as a Bearer token. The request below uses this model's default settings from the catalog and returns a generation id to poll.

curl https://vellria.com/v1/generate/image \
  -H "Authorization: Bearer vll_..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "vellria-minimax-h3-image",
    "prompt": "…",
    "size": "1024*1024"
  }'

Every parameter, the polling call and the upload endpoint are in the API documentation; what the image endpoints allow is on the NSFW image API page.