Become a founding member.

Changelog

Latest product updates

Kling 4.0, Post Analytics and a Team Calendar

Platform
Oct 26, 2026·3.6

Kling 4.0 is on the canvas, and it can edit a clip you upload as well as make a new one. The Video Editor picked up 3D clip transforms and a Gradient Map look. Every model now has its own page, a new Analytics tab shows how your posts are doing, and teams get a shared content calendar. Agents can look up named assets and check a video before anyone watches it.

📦 New

  • Kling 4.0. Make 3 to 30 seconds of video at 720p or 1080p from a prompt, a start frame (with an optional end frame) or up to 10 reference images. Pick 16:9, 9:16, 1:1 or 21:9. It is one card, and what you wire in decides how it runs. See the Kling 4.0 docs.
  • Edit a clip with Kling 4.0. Wire a video into the card’s Source Video port and it edits that clip instead: swap a character, restyle it or change the scene, with up to 10 reference images. Clips run 3 to 30 seconds and need at least 700 pixels on each side. The edit keeps the clip’s length and shape, and the price follows the length of the clip you send. A Start Frame cannot go with a Source Video, so use Reference Images for a look. See Kling 4.0 Edit Video.
  • Seedance 2.0 in 4K for agents and the API. Agents and the API can now set Seedance 2.0 to 4k in its resolution setting. Seedance 2.5 tops out at 1080p today. Its 4K tier is enterprise-only and not in the API yet.
  • 3D and a Gradient Map look in the Video Editor. Select a clip and open its 3D section to set Rotate X, Rotate Y and Z, with the stopwatch to keyframe them. A track can share one camera (Persp., Cam X and Cam Y), so a row of cards tilts together. In the effects list, Gradient Map has a Look menu with Pink negative, Pink duotone, Solarize pink and Hot pink tint. Agents also get text outlines, chroma key, corner pin for a real screen recording on a device, and glitch cuts that carry the cards on top of the scenes.
  • A page for every model. Models lists every model on the canvas, with search and categories. Each model page shows its inputs as real controls (many open with an example filled in) and has the price, an API tab, Copy agent prompt, Download schema and Open in a workflow. The API page now starts from the models too, and both are linked under Docs in the sidebar.
  • Analytics for your posts. The new Analytics tab (it shows once you connect a social account) tracks the Instagram and YouTube posts you published through Wireflow, each against the previous period. YouTube gives views, likes and comments; Instagram adds reach, saves plus shares, and watch time on reels. Instagram connections are in early access. Filter by platform and account, sort by views or saves, and pick 7, 30 or 90 days. Numbers refresh hourly for posts under a week old. Share makes a read-only link that anyone can open without an account for 30 days.
  • A content calendar for teams. Calendar plans your posts by week or month. Make one calendar per client or brand, each with its own color, timezone, connected accounts and brief or brand-board links. A team calendar is visible to everyone on the team: members and admins can plan on it, and viewers can only read. Share lists who has access and copies a link, and you can connect an account without leaving the page.
  • Name your assets. Give a character sheet, a plate, a voice, a brand kit, a recipe or a reference a name, and an agent can fetch it by that name with get_asset. An asset points at the real thing, not a copy, so it stays current. Keep it to yourself or share it with your team (a private voice stays with its owner). Agents read assets with the standard read permission; creating or changing them needs the opt-in write permission.
  • Agent tools for video. check_video watches a clip before a person does and flags a frozen background, lost sound or a black first frame. It costs no credits. get_video_timeline_schema gives an agent the whole Video Editor vocabulary in one read, and edit_video_timeline can now change one clip or add, remove and move clips and tracks without resending the cut (opt-in write permission). create_upload_link takes up to 6 labeled files on one page (also opt-in), and import_social_post copies the slides of a public Instagram or TikTok post for a slideshow. See the MCP docs.

🧠 Improved

  • Every social account in one place. Settings has a new Connected Accounts section with one row per account, a status (OK, Needs reconnect, or a missing permission) and a Reconnect button. You can spot a dead connection before a scheduled post hits it.
  • Chat shows your result. When a run you started from the chat finishes, the video, image or audio appears in the thread with Make another version, Explain this workflow and Show on canvas. A failed run names the step that failed and takes you to it. This is on desktop; the chat-first view already shows results.
  • Shared boards work for the whole team. Anyone who can open a board can now present it. Before, only the owner or a public link could. Editors you share a board with can also read and run it from the Claude connector.
  • Sound stays in sync. A render with a fully silent stretch used to play every sound after it too early (one 20 second video had 18.75 seconds of audio). That is fixed.
  • A new sign-up look. The sign-up questions now build a picture of your first board as you answer: each answer becomes a node wired into an example made on Wireflow. On a phone the Get started button stays on screen.

🔄 Changed

  • Imagen 3, Imagen 4 and Reve are retired. They no longer work, so those nodes now stop with a message before any credits are spent. Saved boards still open. Try Nano Banana 2 or GPT Image 2.5 for the same job.

Wan 3.0, HeyGen Video and an Easier First Run

Platform
Oct 13, 2026·3.5

Wan 3.0 and HeyGen Video joined the video lane, and Seedance 2.5 now renders at 1080p. You can clone a voice with ElevenLabs or HeyGen, Nano Banana 2.1 and FLUX 3 Image landed for pictures, and new video tools can map the motion in a clip or widen a shot to a new shape. New accounts now start on a board they can actually run, and board comments read like a chat.

📦 New

  • New video models. Wan 3.0 and Wan 3.0 Prime make 2 to 30 seconds of video with sound from a prompt, a start frame or reference images, up to 1080p. Prime is the higher quality tier. HeyGen Video makes a 5 to 15 second clip with its own sound and can follow your reference images, clips and audio to keep a product or person the same. Seedance 2.5 now has 1080p in its Resolution menu.
  • New image models. Nano Banana 2.1 edits or combines up to 10 images at 1K, 2K or 4K. FLUX 3 Image comes as a generate node and an edit node that takes up to 10 references, at up to 4K. Ideogram 4.5 Edit stays clean edit after edit and fixes text in place, like a typo on a poster, without changing the font.
  • Clone a voice with ElevenLabs or HeyGen. Upload a clean sample, tick the permission box, and run ElevenLabs Voice Clone or HeyGen Voice Clone. The voice is saved to your account and wires straight into ElevenLabs TTS (Custom Voice) or HeyGen TTS. ElevenLabs v4 also got its own node, with ready-made voices you can search by name.
  • New video tools. Pose Map turns a clip into a skeleton of every person in it, and Depth Map turns it into a grayscale depth clip. Wire either into Seedance 2.5 as a reference video to guide the motion. Video Outpaint keeps every frame and paints new picture around it, so a 16:9 shot can become 9:16. Two Topaz upscalers sharpen finished clips up to 4x.
  • Put an AI influencer in your own video. The new AI influencer page shows Real Swap: bring a vertical clip of up to 30 seconds and it puts the character you pick into it, following the clip’s camera move and timing. You see the cost before you run it.
  • More ways to post. Threads, Bluesky and Mastodon now take text-only posts, and you can cancel a scheduled post before it goes out. With the opt-in social permissions, agents can read a YouTube video’s views, likes and comments at 1 hour, 1 day, 3 days and 7 days after it goes public, or swap the thumbnail on a live video (the channel must be verified for custom thumbnails).
  • Comments that read like a chat. Board comments open in a chat-style panel, full screen on a phone. Reply to any message and it nests right under it. When an agent posts for you, the comment carries an AI badge, so you can tell who wrote what.
  • Agents can do more of what the app does. A new MCP connection can now create workflows and read board comments (reconnect an older one to pick this up). With the opt-in write permission, an agent can also reply to comments, move boards between folders and clear a node’s old results. With the opt-in automations permission, agents can edit or cancel automation slots. Any connection can run your published apps, including new private apps only you can call. wireflow.ai/mcp now has six starter prompts to copy.

🧠 Improved

  • An easier first run. If the goal you type at signup matches no template, it waits in the prompt box with nothing covering it. Your first board starts with a picture step you can run right away, and a one-card tour ends on its Run button with the price shown. "Use this template" keeps your copy through signup. A step your plan cannot run says "Upgrade to run" before you click, and the Run dialog lets you untick it and run the rest.
  • The assistant asks before it replaces a board. If the in-app assistant wants to rebuild a board that already has work on it, it asks first: replace it, or keep it and add next to it. Replace by mistake and one undo brings the old board back.
  • Better motion from a prompt. Describe a video in the Video Editor’s Generate box and the motion it plans is less flat: entrances slow into place and text staggers instead of everything fading in at once.
  • Bigger and longer video. Agents and the API can now upload videos up to 100 MB (it was 25 MB), and Video Editor renders can now run past about five and a half minutes. TikTok Import pulls the 1080p version when the post has one, and Instagram Import works on public reels without cookies.

🔄 Changed

  • Sora 2 is retired. OpenAI removed it from its API on Sep 24, so Sora 2 nodes now stop with a message before any credits are spent.

GPT Image 2.5 and a Voice You Can Actually Reuse

Platform
Sep 29, 2026·3.4

GPT Image 2.5 is live with four nodes, up to 10 reference images on the edit nodes, and five quality tiers. Fish Voice Clone now records a sample right in the browser and saves the voice for reuse. Two new video models and two lyric-capable song generators landed too, and agents can now build motion graphics that the editor will get controls for next.

📦 New

  • GPT Image 2.5. Four nodes: Flare and Sunburst, each as text-to-image and as an edit. The edit nodes take up to 10 reference images, so a real product can land in a generated scene with the label still readable. Five quality tiers from low to max, priced live on the card.
  • Clone a voice once, use it everywhere. Record a sample right in the browser. Clone it with Fish Voice Clone once your consent is checked. It shows up by name in Fish TTS on any workflow after that, no recloning needed. Also reachable through the API and MCP (prepare_voice_clone needs the opt-in write scope).
  • Two new video models. Gemini Omni Flash 1.1 generates from text, a first frame, reference images, or an existing video, at 360p, 720p, 1080p, or 4K. Hailuo 03 Max Turbo renders faster than the clip plays, at half the per-second price of Hailuo 03 Max.
  • Songs with real lyrics. MiniMax Music 2.6 sings section-tagged lyrics you write, like Verse and Chorus. Lyria 3 Pro writes its own lyrics off one prompt and an optional mood image. ElevenLabs Music moved to v2.5 and picked up a lyrics port.
  • Motion graphics, for agents. Ask for a reveal in the Edit composition box, or build one over MCP or the API, and it can now use shape clips (rect, ellipse, line, arrow, path), masks, crop, rounded corners, pop-out cutouts, styled text with a typewriter reveal, and music that ducks under the voiceover on its own. Editor controls for these come next; today they are agent-built.
  • Camera keyframes, in the editor. Stage scenes got real UI for this one: keyframe the camera on its own lanes and move or edit the children directly on the timeline, no scene graph hand-editing needed.
  • New free canvas tools. Color Palette turns colors into a swatch sheet, a prompt string, a list, and JSON at once, free to run. Feed it a mood board or a website’s brand colors, or just type hexes. Painter brushes a mask on an image for regional edits. GLSL Shader runs your own fragment shader over an image and turns its uniforms into sliders and color pickers. Video Trim cuts an exact window for 0 credits.
  • Agents can post for you. New MCP tools, opt-in: list_social_accounts, publish_post, get_post_status. They reach LinkedIn, Threads, Bluesky, Mastodon, and YouTube (YouTube uploads may stay private until Google finishes reviewing Wireflow), with the same ban-safety limits the app enforces.

🧠 Improved

  • Blueprint pages show themselves working. Every public Blueprint page now has a "See it run" gallery of real outputs pulled straight from the flow, so you can judge a blueprint by its results before you fork it.
  • ElevenLabs v4 and v4 Turbo are now options on the ElevenLabs TTS (Custom Voice) node, with inline tags like [whispering] and [chuckles].
  • Instagram Feed format in the Video Editor. Pick Instagram Feed as an output format and set the name, handle, and caption right there.