Runway API v1 — Nano Banana 2 Lite image model:

• New image model nano-banana-2-lite (Gemini 3.1 Flash-Lite Image) on POST /images — Google's fastest image model, up to 14 reference images. Unlimited with exploreMode.
• All Nano Banana models now accept aspect_ratio auto, which is also their new default — the request omits the ratio and the model picks it.

Mureka API v1 — simpler, fully browserless account setup:

• POST /accounts now takes the token and refresh_token you copy from Mureka at sign-in, instead of a Google cookie. The API keeps your session alive on its own — when the token expires it refreshes with no browser and no Google involved. See Setup Mureka.
• New one-page Connect Mureka flow captures both tokens for you — run the snippet, sign in to Mureka, and your account is added.

PixVerse API v2 — Seedream 5.0 Pro image model:

• New image model seedream-5.0-pro (Seedream 5.0 Pro) on POST /images — ByteDance's high-quality generation and editing model. Generates at 1080p (15 credits) or 1440p (30 credits) and accepts up to 10 reference images for image-to-image. Unlimited with Ultra PixVerse subscription.
• seedream-5.0-lite now generates at 2160p (18 credits), alongside its existing 1440p and 1800p tiers.

Runway API v1 — workspace (team) accounts:

• POST /accounts now accepts useWorkspace to bill generation against a paid Runway workspace (team) an account was added to, rather than the account's own plan. Pass true to auto-select the account's single paid workspace, or a team id to pin a specific one. The resolved workspace is returned in the new workspace object and shown by GET /accounts.
• GET /features now returns a teams array listing the workspaces an account belongs to and their plans, so you can find the team id to use with useWorkspace.

Kling API v1 — connect accounts that hit the login slider captcha:

• POST /accounts now accepts an optional passToken + did + userId trio (all three together, or none). Provide them for an account that gets the Kling login slider captcha, which our server cannot solve — the values are your signed-in Kling session, captured from your own browser where you solve the slider once. The account is then stored in passToken mode and refreshes captcha-free (no password login). Accounts that never see the slider still connect with just email and password — nothing changes for them. Setup Kling has a guided browser connect tool that captures the values for you.
• GET /accounts, GET /accounts/<email>, and POST /accounts now return authMode (password or passToken) on every account, so you can tell how each one authenticates.

Mureka API v1 — email + password accounts that stay connected:

• POST /accounts now accepts email + password to connect a dedicated Mureka account. The API stores the password and logs back in automatically whenever Mureka ends the session, so the account keeps working with no manual re-linking — the most durable way to connect. The existing token + refresh_token method still works. See Setup Mureka.
• New guided Connect Mureka with email + password page reveals Mureka's hidden email sign-up, walks you through creating the account, and links it for you.
• GET /accounts and POST /accounts now return authMode on each account — email (email + password, self-healing) or token (session/refresh-token only, re-link when it fully expires) — so you can tell how each one authenticates.

Google Flow API v1 — occasional missing media download links:

Google hands out media download links from an endpoint that limits how often it will answer a given network address. This is Google limiting our outbound calls, not useapi.net limiting you, and nothing you do triggers it. When it happens, Google returns an "unusual traffic" block page instead of a link, and it clears on its own. To put the frequency in context: across the last seven days of our request logs it occurred on a single day, twice — once for about four hours and once for about forty minutes — and not at all on the other six days. It is uncommon, but while an episode is running it affects a large share of link requests rather than a stray one.

• POST /videos, POST /videos/upscale and POST /videos/extend now omit videoUrl, thumbnailUrl, fifeUrl and servingBaseUri when a link cannot be produced, rather than returning one that does not work. Your generation still completed and is unaffected.
• GET /jobs/<jobId> now repairs itself. Poll a completed video job again and any link that could not be produced earlier is resolved and saved, so you get it on a later poll with no code change and no extra call.
• GET /assets/<mediaGenerationId> accepts a new raw parameter. Adding ?raw=true streams the video file itself through useapi.net instead of returning a link, and it keeps working while links are unavailable. It is video only, and it is meant as a way out of a stuck download rather than your normal download path.
• POST /images now documents encodedImage. Whenever an image URL is missing the picture arrives inline as base64 instead, so an image response always carries the bytes one way or the other. Images were never affected by the problem above.

MiniMax API v1 — 768p on Hailuo 3.0:

• Hailuo-3.0 now accepts resolution: "768" alongside 1440 (2K, still the default). 768p is billed at 7 credits per second against 12 for 2K, and a video reference is charged on its own length at the same rate as the output.

PixVerse API v2 — Seedance 2.5 across create, create-frames, and create-fusion:

• seedance-2.5 generates 4–30 seconds in one call, double the Seedance 2.0 family's 15. Two qualities, 480p and 720p — no 1080p or 4K.
• Fusion takes up to 50 references in total — 30 images (frame_1_path…frame_30_path), 10 videos (video_1_path…video_10_path), and 10 audio (audio_1_path…audio_10_path), up from 9 + 3 + 3. Reference clips may run 1.8–30.2 seconds each and summed, against the 2.0 family's 2–15.
• Native audio on every generation. The model takes no audio parameter and rejects one, unlike the 2.0 family's toggle.
• Attaching a fusion reference video lowers the per-second rate — 21 credits/sec at 480p and 45 at 720p, against 35 and 75 without — but the reference's own duration is billed on top of the output. A reference only saves credits while it runs shorter than about two-thirds of your output.
• Available on Pro, Premium, and Ultra plans. Aspect ratios 16:9, 4:3, 1:1, 3:4, 9:16, and 21:9.

Runway API v1 — Seedance 2.5 and Seedance 2.0 Mini on POST /videos/create:

• seedance-2-5 generates 4–30 seconds in one call against the Seedance 2.0 family's 15, and accepts duration: 0 for Auto, letting the model choose the length. Auto bills up to the 30-second ceiling and refunds the unused remainder.
• Up to 50 references in total — 30 images (imageAssetId1…imageAssetId30), 10 videos (videoAssetId…videoAssetId10), and 10 audio (audioAssetId1…audioAssetId10), up from 9 + 3 + 3. Prompts go to 15,000 characters against 3,500.
• Cheaper than Seedance 2.0 at 20 credits/second in 480p and 30 in 720p, against a flat 36. Two qualities only, 480p and 720p, with 720p the default.
• New output_format parameter selects the container, mp4 (default) or mov. aspect_ratio additionally accepts auto, letting the model choose the framing.
• An audio reference no longer needs an image or video alongside it, and each audio clip may run 1.8–30 seconds. Reference videos may run up to 31 seconds each.
• In exploreMode the output is capped at 15 seconds, Auto is unavailable, and the combined duration of all reference videos must also stay within 15 seconds. Credits Mode has the full 30-second range.
• seedance-2-mini is the cheapest of the family at 16 credits/second at either resolution, against 29 for seedance-2-fast and 36 for seedance-2. It offers 480p and 720p, and is otherwise identical to seedance-2-fast except that it defaults to 720p rather than 480p and offers no 1080p.
• Reference-video limits corrected for seedance-2, seedance-2-fast, and seedance-2-mini — each clip may run up to 16 seconds, and the previously documented 15-second combined limit does not exist upstream and is no longer enforced.

Dreamina API v1 — new image model, and per-region availability updates:

• New image model seedream-5.0-pro (Image 5.0 Pro) on POST /images, CA region. Tuned for cinematic and text-rich visuals, with up to 6 reference images and three resolution tiers — 1.5k, 2k (default), and 4k. 1.5k is a new tier no other model offers.
• seedance-2.0-mini now runs on US accounts as well as CA on POST /videos. It previously returned "Model seedance-2.0-mini is not available in region US".
• gpt-image-2 now accepts up to 6 reference images, up from 3.
• On US accounts seedance-1.5-pro and seedance-1.0-mini generate at 720p only, and nano-banana is unavailable. All three are unchanged on CA.

PixVerse API v2 — Relax Mode throttling documented, and the cost calculator corrected against PixVerse's live pricing:

• "Unlimited" Relax Mode image generation is throttled. Past a daily volume threshold PixVerse moves the account into a slow queue instead of refusing the request, so generations still succeed and still cost no credits — they just take far longer to come back. The threshold applies per model, so an account throttled on one model keeps running at full speed on every other one. We have seen the ceiling at roughly 300 a day for nano-banana-pro and 500 for gpt-image-2.0, and not yet on the remaining models.
• queue_data is now documented on GET /images, GET /images/{image_id}, GET /videos, and GET /videos/{video_id}. Alongside queue_count, queue_time, and estimated_gen_time it carries slow_queue and reason, where daily_limit_exceeded identifies a throttled account.
• Third-party discounts are per model, not one rate per plan. Pro and Premium get 40% off seedance-2.0, 10% off seedance-2.0-fast, and 45% off seedance-2.0-mini. Ultra gets 64%, 46%, and 67% on those three, 20% off seedance-2.5, and 40% off every other third-party model. A model-specific promotion replaces the plan-wide Ultra discount rather than stacking with it, and the calculator previously applied a flat 40% to everything on Ultra.
• Seedance 2.0 pricing corrected. PixVerse moved the family to a flat per-quality rate — seedance-2.0 costs 20 credits/sec at 480p, 45 at 720p, 90 at 1080p, and 200 at 2160p, seedance-2.0-fast 15 and 30, and seedance-2.0-mini 8 and 18.
• Yearly billing added to the calculator — Pro $288/yr ($24/mo), Premium $576/yr ($48/mo), Ultra $1,788/yr ($149/mo). Monthly credits are unchanged, so a yearly plan is the same generation at 20–25% less. A fourth plan, Team Ultra, prices identically to Ultra and differs only in offering 12 concurrent generations instead of 8.

PixVerse API v2 — Relax Mode queue behaviour documented, and the completion window extended to match:

• Generations now have 3 hours to complete, up from 30 minutes. PixVerse throttles its unlimited Relax Mode models by moving an account into a slow queue rather than refusing the request, and that queue can hold a generation well beyond half an hour — of those running over 5 minutes, half finish within 18 minutes and one in five takes longer than 30. The window now accommodates it. The expiry message reads 180 minutes rather than 30, so match on the status code rather than the text.
• The daily Relax Mode allowance resets at approximately 00:00 UTC. A throttled account is clear again at the start of the next UTC day.
• Throttling is most frequent on nano-banana-2, then nano-banana-pro and gpt-image-2.0. The point at which it starts varies between accounts, so treat the published figures as rough guidance rather than fixed limits.
• If you have several PixVerse accounts configured, selection now takes the requested model into account. Because the allowance is per model, an account throttled on one model stays fast on every other one, so requests are steered towards accounts with nothing already queued on the model you asked for. This can only use headroom that exists — an account pool that is throttled on a model throughout has no spare capacity to find, and the answer there is more accounts or a different model.