Sign in

Changelog

Model and API changes, with dates.

Model and API changes, with their dates. The newest is first. A retirement of a model is announced here at least 14 days before it happens.

Wan 2.6 retired

2026-10-28 · Model change

Wan 2.6 (wan/2-6) is retired today. It is no longer in the model list, GET /v1/models, and a task that names it now answers not_found, with no charge. Tasks made on it before today still read, with their clips, until they expire. Wan 2.2 Adult (wan/2-2-adult) stays. Veo 3.1 Lite (veo/3-1-lite) and Hailuo 3 (hailuo/3) still make a clip from the prompt alone. This retirement takes effect at once, without the 14 days' notice described above.

Video: a last frame on Veo 3.1 Lite and Hailuo 3

2026-10-24 · API change

A video task may now give the picture its clip ends on, in last_frame, beside the first frame in image. Veo 3.1 Lite and Hailuo 3 take one. Wan 2.6, Wan 2.2 Adult and H3 Image to Video do not.

  • The model list says which: limits.video.last_frame is true or false.
  • Send last_frame together with image. A last frame is checked and held to the same limits as the first frame.
  • The price is the same as without a last frame.
  • A last_frame on a model that takes none, or without image, is 400 unsupported_combination, with param set to input.last_frame. Nothing is charged.

Nano Banana Edit and FLUX 3 Image Klein retired

2026-10-21 · Model change

Two models are retired today: Nano Banana Edit (nano-banana/edit) and FLUX 3 Image Klein (flux/3-klein). They are no longer in the model list, GET /v1/models, and a task that names one now answers not_found, with no charge. Tasks made on them before today still read, with their results, until they expire. FLUX.2 Klein (flux/2-klein) was retired on 2026-10-17, so no Klein model is left. Qwen-Image-Edit 2511 (qwen-image-edit/2511) stays in the model list as an edit model, GET /v1/models?task=edit. Nano Banana 2 (nano-banana/2) and Nano Banana 2 Lite (nano-banana/2-lite) stay, and each still takes one reference image. This retirement takes effect at once, without the 14 days' notice described above.

Video opens: five models, paid by the second

2026-10-21 · API change

Dramatis Studio makes short videos, and the five video models are open to every account. A video task is created with the same POST /v1/tasks, with the task kind image_to_video: a prompt, a duration, and a first frame in image. A clip takes minutes, so poll the task or take a webhook; the Video guide shows both, with samples in cURL, JavaScript and Python.

The models are in the model list, GET /v1/models?task=image_to_video, each with its own model page:

  • Veo 3.1 Lite (veo/3-1-lite), Wan 2.6 (wan/2-6) and Hailuo 3 (hailuo/3). They also make a clip from the prompt alone: leave image out, at the same price. They may refuse a request.
  • Wan 2.2 Adult (wan/2-2-adult) and H3 Image to Video (h3/i2v), which are labelled adult-capable and need a first frame. A task with none is 400 unsupported_combination, with param set to input.image, and nothing is charged.

Veo 3.1 Lite is live; the other four are in beta. What is new in the contract:

  • output.video on a finished video task, in place of output.images. It holds the link to the clip, the poster, and the length, frame rate, size and sound measured from the file. The event of a webhook carries it too. A clip made from the prompt alone has no poster: output.video.poster is null.
  • limits.video on a model: the lengths, aspect ratios, sizes, frame rate, sound, whether the first frame is optional or required, and the largest file. It is present on a video model only. Wan 2.6 and Hailuo 3 keep the shape of the first frame, so they list no aspect ratio and no size; a clip they make from the prompt alone is wide (16:9). Hailuo 3 may return sound, which cannot be turned off.
  • pricing.unit on a model, image or second. A video model is paid by the second you ask for, and its quote has one line, second of video. output.video.duration_seconds is the length of the file, which can differ a little.
  • A video task has running limits of its own, apart from the image limits. A failed video is refunded in full.

Nothing changes for an image task.

Three Krea 2 models added, FLUX.2 Klein retired

2026-10-17 · Model change

Three new models are in the model list, GET /v1/models: Krea 2 Large (krea/2-large), Krea 2 Medium (krea/2-medium) and Krea 2 Medium Turbo (krea/2-medium-turbo), image models from Krea. Medium is the balanced one, Medium Turbo is distilled for speed, and Large is the larger model with a rawer, more textured look. They are in beta. Each makes images at 1k in the ratios 1:1, 2:3 and 3:2, and takes no reference image. Large can take a minute or more for one image. FLUX.2 Klein (flux/2-klein) is retired today: it is no longer in the model list, and a task that names it now answers not_found, with no charge. Tasks made on it before today still read, with their results, until they expire. FLUX 3 Image Klein (flux/3-klein) stays. This retirement takes effect at once, without the 14 days' notice described above.

Editing opens with two models

2026-10-16 · Model change

Two edit models are in the model list, GET /v1/models?task=edit: Nano Banana Edit (nano-banana/edit) and Qwen-Image-Edit 2511 (qwen-image-edit/2511). An edit task starts from your picture: give image (a link, a small data: image or an upload) and a prompt that says what to change. Both are in beta and make one png at 1k. aspect_ratio defaults to auto, which keeps the proportions of your picture. Nano Banana Edit takes up to three further images in images; Qwen-Image-Edit 2511 takes none and also takes a seed. There are no masks. Each has its own model page with the form, an image picker and a before-and-after view. A picture can be 65,536 to 16,777,216 pixels and up to 10 MB.

Seven more models retired

2026-10-16 · Model change

Seven more models are retired today: UnNamedA Anima Realistic (anima/unnameda-realistic), Sam Anima Realistic (anima/sam-realistic), A-real Anima (anima/areal-realistic), Anima Turbo Character Reference (anima/turbo), Seedream 4.5 (seedream/4-5), GPT-5 Image Mini (gpt-image/5-mini) and Riverflow V2 Fast (riverflow/v2-fast). They are no longer in the model list, GET /v1/models, and a task that names one now answers not_found, with no charge. Tasks made on them before today still read, with their results, until they expire. This retirement takes effect at once, without the 14 days' notice described above.

Anima Character Reference (anima/character) is the one Anima model that stays, and nothing about it changes. The other models of the model list stay as they were.

Wide and tall sizes added to Z-Image Turbo by Stable Yogi

2026-10-16 · Model change

Z-Image Turbo by Stable Yogi (z-image/turbo-stable-yogi) now makes images at 1k in the ratios 3:2 (1216 x 832) and 2:3 (832 x 1216), beside 1:1. The price does not change: it is set by the resolution, and 1k covers every ratio. The other Z-Image model, Z-Image Turbo Anime, still makes 1:1 only.

FLUX 3 Image Klein added

2026-10-16 · Model change

A new model is in the model list, GET /v1/models: FLUX 3 Image Klein (flux/3-klein), Black Forest Labs' smaller, faster FLUX 3 image model. It is in beta. It makes images at 1k in the ratios 1:1, 2:3 and 3:2, and square images at 2k, and takes one reference image. FLUX.2 Klein (flux/2-klein) stays. A request for a ratio or a tier it does not list answers unsupported_combination, as for every model.

Seven Illustrious models retired

2026-10-16 · Model change

Seven of the eight Illustrious models are retired today: Illustrious Anime RC1, Anime RC2, Anime Adult, Anime Adult RC1, Realistic, Realistic RC1 and Realistic RC2. They are no longer in the model list, GET /v1/models, and a task that names one now answers not_found, with no charge. Tasks made on them before today still read, with their results, until they expire. This retirement takes effect at once, without the 14 days' notice described above.

Illustrious Anime (illustrious/anime) stays. It is now marked as able to make adult images, as the label on its page says; nothing else about it changes.

First release: API v1 and the first models

2026-10-15 · API change

Dramatis Studio opens with version v1 of the API. Samples use https://api.studio.dramatis.ai/v1, which is not open yet. The version is in the path, and additive changes do not bump it.

The API has tasks (create, read, list, cancel, delete content), quotes, uploads, the model list, the balance, the usage and the calling key. Results can be pushed to your server by signed webhooks. The TypeScript and Python packages, both to be named opsis-sdk, are not published yet. The docs and the API reference describe the calls.

The models on offer are:

  • Illustrious: Anime, Anime RC1, Anime RC2, Anime Adult, Anime Adult RC1, Realistic, Realistic RC1 and Realistic RC2.
  • Z-Image Turbo: Anime, and by Stable Yogi.
  • Anima: Character Reference, Turbo Character Reference, UnNamedA Realistic, Sam Anima Realistic and A-real Anima.
  • Provider models: Nano Banana 2, Nano Banana 2 Lite, Riverflow V2 Fast, FLUX.2 Klein, Seedream 4.5, GPT-5 Image Mini and Qwen Image 3.

The model list, GET /v1/models, is the record of every model, its inputs, its limits and its prices.