AI Model Updates: Image, Video, Audio and Tools | SnapGen

New creative models, supported settings and platform updates, newest first. Open a model page for its current availability and price.
Create a new image from a prompt, or use up to ten reference images to guide an edit. Choose 1K, 1.5K or 2K output and a supported aspect ratio. Each reference image adds to the request price; check the total before you create.
Explore Seedream 5.5.
Create 3–15 second videos with sound from text, a first and optional last frame, or up to seven reference images. Choose 360p, 720p, 1080p or 4K; first and last frame generation supports up to 10 seconds.
Explore the model and current prices.
Kuaishou Kling 4.0 is available in three modes on POST /v1/videos: kling-4-t2v for text, kling-4-i2v for a first and optional last frame, and kling-4-r2v for one to nine image references.
Request any whole-second duration from 3 to 30, at 720p or 1080p, with native stereo audio. Text and reference modes support 16:9, 9:16, 1:1 and 21:9; image-to-video follows the first image. This release does not expose source-video editing, intermediate keyframes, 4K, HDR or extension.
Billing follows the requested output seconds and resolution. Submit, poll the video task, then download the completed MP4.
Open the model Playground and live pricing.
OpenAI GPT Image 2.6 is available in Flare and Sunburst versions on POST /v1/images/generations.
Use gpt-image-2.6 for Flare by default, or select gpt-image-2.6-sunburst directly. Each request returns one 1K image, with six quality choices and up to sixteen references. All supported quality settings and both versions use the same per-image billing; references have no surcharge.
This release returns PNG, JPEG or WebP. SVG, GIF, transparency controls, masks and markup editing are not exposed. A successful waiting request returns image URLs; Prefer: respond-async returns a task to poll.
Open the model Playground and live pricing.
Meta Muse Video is available through muse-video-t2v and muse-video-r2v on POST /v1/videos.
Each request generates one 720p clip of about ten seconds with native sound. Reference mode accepts one to nine images, with an 11 MiB combined original-file limit. References guide the scene; they are not fixed first or last frames.
Both modes bill per request. The current route does not expose alternate durations, silent output, input video/audio, editing or extension. Poll the returned task and save its completed MP4 promptly.
Open the model Playground and live pricing.
Nano Banana 2.1 is available through POST /v1/images/generations with the public model ID nano-banana-2.1.
Generate a new image from a prompt, or supply up to 16 reference images to guide edits. Each request delivers one image at 1K, 2K or 4K. Set size to auto or one of ten aspect ratios and n to 1.
Billing is per delivered image at the selected resolution. Reference images have no additional charge. A successful waiting request returns data[].url; send Prefer: respond-async to receive a task immediately, then poll GET /v1/tasks/{id} with the same API key. Download successful outputs promptly because result URLs are temporary.
Try the Playground and compare live resolution prices on the Nano Banana 2.1 model page, or follow the Nano Banana 2.1 API guide.
The GPT text lineup now starts at GPT-5.6 and includes seven model IDs:
Keep your existing SnapGen API key, prepaid balance and base URL, https://api.snapgen.org/v1. All seven IDs are available on POST /v1/chat/completions and POST /v1/responses; set stream: true for streaming replies.
GPT text models below 5.6 are no longer offered. Update older model IDs in your application to one of the available variants. The unsuffixed gpt-5.6 ID remains unavailable: use Luna, Terra or Sol explicitly.
The three GPT-5.6 tiers keep their existing rates. Current input, output, cached-input and long-context rates are on the GPT-5.6 and GPT-6 model pages.
Other text model families, including Gemini, Grok and DeepSeek, keep their current availability. Claude's existing availability is unchanged, as are image, video and audio models. See the model catalog and Chat Completions reference for model IDs and request details.
Six ElevenLabs model families are live on endpoints under /v1/audio/, all on your SnapGen API key:
Every family except dubbing answers in the same request, so there is nothing to poll. SnapGen measures the length of uploaded audio itself, and failed requests are not charged.
Compare the six families and their live prices on the ElevenLabs provider page. In the docs, start with the text to speech reference.
Suno V6 and Flow Music, which runs Google's Lyria 3 Pro and Lyria 3.5, are live on SnapGen. Both use POST /v1/music/generations: each call starts a task that you poll with GET /v1/tasks/{id}, and a finished task can be passed as source_task_id to edit its track.
Every operation is billed per request, and failed generations are not charged. Result links last about 72 hours, so download what you want to keep. On each model page, the audio Playground has a tab per operation, and Continue with… sends a finished track to the next one.
Live prices are on the Suno and Flow Music model pages, and the music generation reference lists every operation.
Gemini Omni Flash 1.1, Google's fast video model, is live in four modes on POST /v1/videos:
Clips run 4, 6, 8 or 10 seconds with native audio, at 360p, 720p, 1080p or 4K in 16:9 or 9:16. Generation is billed per output second by resolution, and extend is billed per request. Jobs run as tasks: submit, poll the status, then download the MP4.
Compared with Gemini Omni, which makes fixed 10-second clips at 720p or 1080p, Flash 1.1 lets you choose the length, adds 360p and 4K, and takes more reference images.
Try it in the Playground on the Gemini Omni Flash 1.1 model page, or read the API guide and the video generation reference.
Seedance 2.5 gains four mode models next to the 480p and 720p SKUs that launched in August. All use POST /v1/videos:
Clips run 4 to 30 seconds with opti