What's new
image-genanime
- The model is fine-tuned on a new dataset version: poses, anatomy and complex multi-person scenes come out more accurately.
- You can now choose the quality: Medium — up to 2 MP, as before, and High — up to 3 MP for large, detailed shots.
- A new upscaler between generation stages preserves fine details better when increasing resolution.
- The grid artifacts that sometimes showed through on smooth areas like skin and sky are now removed automatically.
- Velvet Render v5.1 Anime gets the update too.
image-editor
- The editor now includes the new ImageGPT v3.5 model with improved understanding of complex instructions and better preservation of character identity, object details, and text.
- It supports up to 8 reference images. Refer to them as
<image1>,<image2>, and so on; the first image determines the output proportions. - A single run can produce up to 4 image variants at resolutions of up to 4 megapixels.
- The model is suitable for replacing clothing and objects, transferring styles, working with multiple characters, product imagery, posters, and lettering.
image-genimage-editor
- OpenAI Image 2 has been replaced by version 2.5 in the image generation and editing services.
- The model produces sharper details, more natural lighting, and more realistic textures.
- When working with references, it better preserves faces, distinctive character features, and the composition of the source image.
- Editing is more precise: the model is better at changing only the requested elements while preserving everything else.
- According to OpenAI, the new version follows complex visual instructions more reliably and runs up to 50% faster than its predecessor.
- It is available at 1K resolution with two quality levels: Standard and High.
image-enhance
- Pic Refiner 2.1 handles lighting noticeably better: light looks more natural and dimensional, while shadows, highlights, and reflections are more realistic.
- The rendering of skin, hair, fabric, foliage, and other complex materials has been improved.
- The model preserves the composition and details of the source image more accurately, producing a more cohesive and expressive result.
animeimage-gen
- Animake v1.1 now supports two styles: Smooth Lines and PuAnime.
- Choose a style before generation or keep the model's default look.
animeimage-gen
- Added Animake v1, a fast model for creating anime characters, illustrations, and artistic scenes.
- It generates images quickly, follows the selected style consistently, and works well for iterating on prompts.
- Prompts are best written in English, but other languages are also supported: text is translated into English automatically.
- Supports natural-language descriptions, Danbooru tags, and combinations of both.
- Each task can generate up to 8 images.
- Each image costs /bin/zsh.03.
music-gen
- BeatNova v1 has been added to the music generation service for creating complete songs and instrumental compositions.
- You can start with a short track idea: the service will automatically prepare a detailed prompt and structured lyrics. You can edit the result before generation.
- BeatNova v1 maintains the style, melodic themes, vocal character, and arrangement development throughout the composition.
- The model creates expressive vocals and supports complex track structures, including intros, verses, choruses, bridges, and instrumental sections.
image-enhance
- We nearly halved the cost of enhancing image quality with Hyper Upscale v4: from $0.11 to $0.06.
image-gen
- Prices have been reduced for MidJourney v8.2, v8.1, and v7.
- Four images now cost $0.0675 at 1K and $0.108 at 2K.
- 2K resolution is available for MidJourney v8.2 and v8.1.
video-gen
- Reduced video generation prices for MomentFlow v5.2 across all available resolutions, both with and without a video reference.
- Also added support for 1080p resolution.
Page 1 of 11