What's new on Layer

Hand the Layer agent a Photoshop file and get back a character rigged and animated for Spine. Also this week: textures that repeat without a seam, a brand you can carry in a Reference Set, and our regular refresh of the models behind your everyday tools, so you get the best results without having to pick.

What's new?

Rig a Photoshop file for Spine: Give the agent a layered Photoshop file and ask it to rig the character, creature or prop for Spine. It reads your layers as parts, builds a skeleton with bones and pivots, and animates an idle plus up to three clips you ask for, for Spine 4.1, 4.2 or 4.3. No layers? A flat image works too: Layer splits it into parts first. Every animation plays in the viewer before you export, so you can ask for changes in the same chat until the motion is right. New this week, a rig goes beyond the bones: a surface can breathe, squash or ripple; a boot or a glove declares its opening, so the limb shows inside the cuff instead of behind the painted rim; a character facing you gets a walk that steps in place; and each clip can carry its own face. A rig needs at least 200 Creative Units in the workspace to start.

The best models, without the homework: New models arrive every week, and we keep testing them so you don't have to. This week's refresh moves the defaults for a new image, an edit, upscaling, Create 3D, the 3D composer, retexturing and splitting a mesh into parts onto the models that do each job best. Open a tool and it is already set up; nothing to compare, nothing to change. GPT Image 2.5 also costs less than it did. You can still pick any model yourself, and the models page lists every one.

Tileable textures: Ask for a texture that repeats, and the seams are joined. On models that offer it, Tileable is in the advanced settings. You can also ask the agent, including on a model whose form has no switch. Pixel art and organic surfaces are joined without another generation. A structured texture, bricks or planks, still repaints the seam, and that step spends Creative Units. From a texture you already have, the agent can derive a normal map, a height map, ambient occlusion, roughness and emissive that tile with it. Those maps do not spend Creative Units. Pixel art is worked one cell at a time.

A brand, as a Reference Set: A Reference Set can carry a brand: a palette of up to 16 named colours, logo marks, a wordmark and fonts. Whenever you use the set, the palette is added to the prompt, except on audio models. Logos and fonts stay on the set. The agent places a logo when you ask for it, and it does not train on those files.

Insert a scene: On a video, Insert scene marks the stretch to replace. MiniMax H3 Max Insert writes a new scene into the clip, and you get the whole clip back. You spend Creative Units for the new scene. The clip you get back includes the original footage around it. A scene runs from 5 to 13 seconds, at 480p or 768p, and the stretch you replace has to begin more than a second and a half into the clip.

Clean up a recording: Denoise is available on an audio file, from the same place you denoise an image. VEED Clean Audio is the default: it keeps the speech, including quiet and distant words, at the same length as the source. It sits with the rest of audio cleanup.

Exact words, in your font: Ask the agent to set text in a font from your workspace. The words are rendered exactly, on a transparent background, ready to place on an image. That step does not spend Creative Units.

A workflow can drop weak results: A workflow can score every item in a list against a rubric you write, and pass on only what clears the bar. There is a judge for text, images, layered images, video and meshes. If nothing passes, the step stops and says so.

Layer Plugins are live, in seven places: Photoshop, Premiere Pro, After Effects, Blender, Figma, Slack, and Chrome. Each one puts Layer where the work already happens: generate against the document, the project, the composition, the scene, the canvas, or the thread in front of you, and the result lands back in it. Every plugin is free with a Layer account, signs in with the account you use on the web, and generations spend Creative Units the same way they do in the browser. Everything you generate is saved to a Layer session, so your team finds it where it always looks.

Layer Plugins are Live

Photoshop: The Layer panel for Photoshop generates against the document you have open, on Photoshop 25.10 or later, on macOS and Windows. Your selection is the mask: marquee, lasso, or Select Subject, and only that area is regenerated, placed back as a new layer with a layer mask, feathered so it sits on the image. Inpaint, relight, change the camera angle, extend the canvas (a reframe of the whole document grows it, as its own undo step), split a flat image into layers, or remove a background, with the full model catalogue and your trained styles behind one prompt box. The panel is redesigned this week: it uses Photoshop's own type, can take a style-transfer prompt, suggests Reference Sets from the project you are in, says when a capture is read smaller than the document, and tells you when an update is out. Install it from the .ccx on the page with the Creative Cloud desktop app.

Premiere Pro: The Layer panel for Premiere Pro generates video and audio beside the edit you are cutting, on Premiere Pro 25.6 or later, on macOS and Windows. Describe a shot or a sound, and the result is imported into your project, ready to drag onto the timeline. It installs from a .ccx, the same way the Photoshop panel does.

After Effects: The Layer panel for After Effects puts generation next to the timeline, on After Effects 25.0 or later, on macOS and Windows. Describe a plate, an element, an end-card badge, or a sound, and the result arrives as footage at the top of the layer stack, sized to the region it was made for, in one undo step. Image tools work from the frame you are on. Footage the panel placed can be selected and changed with a prompt, reframed for a vertical placement, or upscaled. For any other clip, the panel asks you to generate one here or add it from your Layer session, then select that layer. It docks beside Effect Controls, and generation runs on Layer rather than on your machine, so a render never competes with it for the same cores. Install it from the .zxp on the page with a ZXP installer.

Blender: Layer for Blender is a native add-on for Blender 4.5 or later, on macOS, Windows, and Linux. Its tools and its agent sit in the 3D viewport and work on the scene you have open: meshes, textures, images, and video land in the file, and a change the agent makes is one undo. Restyle and motion to video can use the scene's animation. The add-on opens on the prompt bar, keeps the agent's history, updates itself, and says when a restart is due.

Slack: Layer for Slack makes a thread a session. A workspace admin connects it once. Mention Layer in a channel or a direct message, and that thread gets the same agent, your styles and Reference Sets, with the reply showing as it is written. Images, video, and audio are posted back into the thread, and a 3D model arrives as its rendered preview. A prompt sent from Layer into the thread names who sent it, and the thread links to the session, where the full files live.

Figma: The Layer plugin for Figma runs from any file. Generate backgrounds, icons, buttons, and characters in the panel and place them straight onto the canvas, with no export and re-import. Install it from the Figma Community.

Chrome: The Layer Reference Collector saves reference images from any website into your Layer workspace: one click, or a batch selection of many images at once, with optional auto-upload. It works in Chrome and Edge, and it is free with any Layer plan.

What's new?

Spine, from a PSD or a flat image: Ask Layer to rig a character, creature, or prop for Spine. It starts from a layered PSD, or from a flat image that it splits into parts, and it asks you to confirm the Spine version (4.1, 4.2, or 4.3), the clips, and the cost before it runs. A rig is a longer job: it needs at least 200 Creative Units in the workspace to start. The result is a skeleton with bones and pivots, an idle, and up to three clips you asked for, and you preview it right in Layer: the rig and every animation play in the viewer, whichever Spine version you chose, so you can check the motion, ask for changes in the same chat, and iterate before you export. Parts can bend as weighted meshes instead of hinging as rigid cutouts, and on Spine 4.2 and 4.3 hanging pieces such as hair, capes, and tails can carry physics.

350+ cinematic presets: A new library of Reference Sets names a technique so you do not have to describe it from scratch, each with a cover that shows it in action. Mention one in a prompt, or pick it from the video form, where they are grouped as Camera, Staging, Look, and Performance. There are twelve types:

• Camera Movement: dolly, crane, orbit, whip pan, and more
• Camera Angle
• Camera Framing
• Camera Lens
• Scene Lighting
• Scene Composition
• Scene Atmosphere
• Genre Look
• Optical Effect
• Visual Effect
• Character Performance
• Time and Motion: speed ramps, slow motion, and other changes of pace

Moves and changes of pace are for video. Every other type describes a single frame as well, so it works on image models too.

Extend and speech: Extend video defaults to MiniMax H3 Max Extend. You get the whole clip back, the source plus the new seconds. Gemini 3.8 Flash TTS is the default for speech.

Music, section by section: On audio models that take a composition plan, the advanced panel is a list of sections: what holds for the whole track, then each section's length, what enters, what stays out, and any sung lines. The same plan can be sent from the API and from MCP.

OBJ and FBX downloads: A 3D model can be downloaded as OBJ with its textures as PNG files, without spending Creative Units. Tripo models can also be downloaded as FBX, with the rig and the animation kept. That conversion spends Creative Units.

Several images, one Photoshop file: Exporting a selection of images as PSD can combine them into a single layered file: one layer per image, the first you selected on top. A PSD that already has layers keeps them as a group.

Upscale from the composer: Upscale an image or a video you upload, from the same composer you generate in. Image upscalers that support it expose creativity and resemblance there. Leave them unset and the model keeps its own defaults.

Save the session as a template: Save as template now keeps the session: its name, and the generation settings that were on the form. You can edit those settings later, instead of recapturing the session.

A rule in chat can become a skill: State a rule the work should follow, and the agent can save it as a Workspace Skill on the project you are in. It confirms what it will write, and what it will be called, before it saves, unless you already named the skill and the change. Later chats in that project follow it.

An animated mesh plays: A 3D model that carries an animation gets a short, silent, looping preview. It plays in the viewer and on the session tile, the same way a video does.

This week brings Qwen Image 2.1 and new defaults for lip sync and video reframe, webhooks so an API call can hand back its result without polling, and an optional onboarding conversation that can build a first project with you.

What's new?

Qwen Image 2.1: Qwen Image 2.1 is the latest image model from the Qwen team at Alibaba, and one of the strongest open-weight models you can run today. It is built for prompt adherence, so a long brief lands closer to what you asked for, and it generates with control over size, guidance, seed, negative prompt and transparency. Qwen Image 2.1 Edit takes the same model to prompt-guided editing, combining or transforming up to 10 reference images, so a scene can be rebuilt from art you already have.

Video reframe and lip sync: LTX Video 2.3 Quality Outpaint is the default when you reframe a clip. MiniMax H3 Max Lip Sync is the default for lip sync, from a still plus a voice track, up to 2K.

Onboarding can start with the agent: A new user with an active workspace who has not answered onboarding can be offered an agent-led setup: it asks what you are working on, pulls your game's art in from its store page, and makes something with you. Taking it is optional, and that first conversation does not spend Creative Units. We are trying a few different welcomes, so not every new user meets the same one.

The agent leaves beta: The Agent tab no longer says Beta. Most agent actions still spend no Creative Units; the intensive ones do.

Webhooks: An API call can name a URL and receive the finished result there, signed, instead of polling. The response lists the deliveries to expect: the job's own outcome, plus a separate scoring event where the workspace scores automatically. A test call sends a signed sample to any URL, so you can check your verification before spending Creative Units. The webhooks guide has the payload, the signature headers, and what happens when your endpoint is down.

The project stays with you: The project you are in is one setting, carried across the app. Search, listings, and sessions stay inside it, and search finds session templates too.

A shared workflow runs: Opening a shared workflow adds it to your workspace as you run it. You no longer have to link it first. Publishing is three steps (Inputs, Outputs, Publish), with names and examples edited on the card.

Editable type in a Photoshop file: When a layered result carries its text, whether a layout the agent measured or a PSD that arrived with type layers, the copy comes back as live type in the face it was set in, rather than flattened into pixels.

Find a teammate: Workspace members can be searched by name or email, filtered by status, and exported. Built for studios with large teams.

What's new?

Layer from the terminal: The Layer CLI and TypeScript SDK put generation, editing, and training where your scripts already are — one login, the same models and Reference Sets as the app. The agent skills teach a coding agent how to drive them.

Masked edits on Gemini: A masked edit is sent as an annotated init image, so Gemini sees the region you marked rather than inferring it from the prompt alone.

Command palette, with room to work: ⌘K opens a larger window with standard chrome, so searching projects, sessions, files, and models isn't cramped.

Arrow keys page the viewer in gallery order — next and previous without leaving the asset.

Exports name what they produced: A finished row is labeled with the container it wrote (zip, sequence, and so on). Exporting a video as a PNG sequence extracts the frames.

Usage, in more detail: Settings → Usage again shows the subscription and purchase detail. Chat agent spend is listed per agent, with readable token pricing.

What's new?

GPT Image 2.5 — Flare and Sunburst: OpenAI's GPT Image 2.5 is in the catalog for text-to-image and edit. Flare is now the top image-editing pick, ahead of GPT Image 2 and Gemini Flash Image.

FLUX.3 Edit: Prompt-based video editing on Black Forest Labs' FLUX.3 — describe the change, keep the shot.

MiniMax H3 Max Turbo: A faster H3 Max for text-to-video and image-to-video at 480p, 768p, and 1080p, when you want the same model family at draft speed.

Meshy T2: Meshy's flow-matching native mesh generator — text or a single image to 3D, with textures and PBR. On Meshy V7 and T2 you can now request ultra detail, remesh, and a texture image.

Camera moves on H3 Max: Author an orbit path with 2–12 keyframes (azimuth, elevation, distance over time) and H3 Max flies it. From the Camera move tool on an image, or from the generation form when the model can take a trajectory. A thumbnail of the path sits on the control once it's set.

Favorites across the board, gallery, and library: One heart on the tile. Filter a session to Favorites, or Group by favorites (⇧V) to collect them into a board group everyone in the session can see. The library sidebar has a Favorites view for what the workspace has picked out.

Video editor, from the clip: Open the editor from a video — full screen, with a cover on the tile — instead of assembling only from a blank timeline.

Several source images on edit models: Attach more than one image when the model can take them; generations fan out from the set.

Skills from the project: Create or import a skill in the project picker without leaving the page. Open a skill from the project sidebar. Project instructions convert into project-attached skills.

Asset type filter on the session board and gallery, replacing the items-panel type row.

Del and ⇧Del on the board: Del clears a failed generation; ⇧Del deletes the asset.

Timed votes on canvas objects, with stamps that outlive the round — the timer is only a timer.

QR code in Export: The lightbox Export section shows a QR code for the asset URL.

Session Creative Units in chat: The rail shows Creative Units used in the session.

Provenance on exports: C2PA claims are signed at file creation and on re-encoded exports, including video and audio.

A full-screen asset viewer again, from the session.

Retrain a Reference Set when a finetune predates the set's current files.

Edit Image Auto size matches the base image, the same Auto the composer already offered.

Downloads run over the websocket, so a file transfer no longer depends on a separate HTTP path.

Exports say what they're doing while they run, and they outlast a connection blip instead of failing silently.

What's new?

MiniMax H3 Max — top-ranked video, with the sound already in it: MiniMax's post-trained H3 is now in the catalog, and it is currently at the top of the blind-preference video leaderboards (Design Arena and Artificial Analysis). What it changes for a game team:

• Audio is generated with the picture, not dubbed on after. Footsteps, impacts, ambience, and dialogue land on the frames they belong to, in native stereo — so a trailer beat or a cutscene is watchable straight out of the generation.
• It hits the beats in the order you wrote them. Prompt adherence is the model's headline improvement over base H3: a multi-action shot list plays back as a shot list, not as a vibe.
• Camera moves and character look hold across a shot. Directed pushes, tracks, and pans, with a character and an art direction that stay themselves from one cut to the next — the part that usually forces a re-roll.
• Ship-ready range. Text-to-video and image-to-video, 5–15 seconds, 480p, 768p, 2K, or 4K, six aspect ratios from 21:9 down to 9:16, and optional first-to-last-frame control for transitions you actually specify. Draft at 480p, then run the keeper at 4K.

Comments on the session board: Press C or use the toolbar to pin a thread on a tile or on the board itself. Reply, @mention teammates, resolve, and filter resolved threads — pins follow their tile as you rearrange the canvas.

Live presence and cursors: See who's in the session from the avatar stack in the header, and follow teammates' cursors on the board in real time.

Recraft V4 custom styles: Train a Recraft style from your reference images and generate with it in both raster and vector, on Standard or Pro. One training run covers both output kinds of that tier.

Background exports: Queuing a download no longer holds the dialog open. Exports run in the transfers panel alongside uploads, keep going if you close the dialog, and can be downloaded again from the finished row.

Auto-reload Creative Units: Opt in per workspace to top up automatically when the balance drops below a threshold you set, so a long session doesn't stall mid-generation. Enterprise and child workspaces stay on their existing billing.

A bigger asset viewer, without leaving the session: Opening an asset expands it into the session canvas instead of covering the whole page — header and chat stay put. In the viewer, both rails resize, the pager sits on the stage, and you can zoom to pixels.

Command palette finds everything: ⌘K searches projects, sessions, files, reference sets, workflows, and models in one ranked list. Hits on files, sets, and workflows open as shareable filtered URLs.

Video editor, faster to assemble: A CapCut-style rail for Image, Video, Audio, and Transition; right-click the timeline to add a clip at that frame; pick assets from browse and they land on a new track at the playhead. After a render, the board focuses the output.

Workflows: Segment Mesh is a node (the last 3D editor op that was missing one). And when a draft is better than the published original, Publish as new workflow forks it instead of overwriting.

Reference sets: A strength meter on the set editor shows how ready the set is to train. Artists can now create and manage sets without the training permission. Project tiles for workflows and sets offer Edit in place.

Auto aspect ratio: Models that can take their output size from the source image or first frame now offer Auto in the size picker, so image-to-video follows the source instead of snapping to a preset.

Home composer follows the file you attach: Drop a source video, mesh, or audio onto the matching tab and the composer switches to a model that can actually read it — with a one-click undo if you wanted the original model.

What's new?

Lip sync from the lightbox: Drive a talking clip from a video plus an audio track, or animate a still into speech, without leaving the asset viewer. Sync.so, LatentSync, and VEED models are all available from the same control.

Video editor on the session canvas: You can now drop a video editor onto the session board and assemble clips, titles, and audio there — the same timeline that used to live only in workflows, available wherever you already work.

Interpolate and slow motion: Frame interpolation is its own tool. Convert frame rate or slow a clip down by up to 8× with Topaz Apollo, Chronos, and Aion, from the lightbox, the editor, or a workflow node.

The full Topaz lineup: Dozens of Topaz models are now available across image and video. On the image side, that's the Gigapixel precision tiers (Standard V2, High Fidelity V3, Low Resolution V2, Standard MAX), the generative ones (Redefine, Recover, Bloom, Wonder), plus CGI, Text Refine, and Transparency Upscale for game-art edge cases — with Nyx, Denoise, and Dust-Scratch for cleanup. On the video side: Proteus, Iris, Dione, Artemis, Gaia, Rhea, Theia, Astra, the Starlight family, and SDR-to-HDR.

Score assets against studio standards: A workspace or project can describe what “on-standard” looks like — in words and with reference images — and finished assets are scored 0–100 with a rationale you can act on. Sort and filter the library by score, and re-score a library when the rules change.

Input guardrails: The same policy layer can screen a prompt and its input images before a generation runs, either blocking the request or warning and letting it through. Workspace rules always apply; a project can only add to them.

Layered image editor: Undo and redo in the layered viewer, plus lock, group/ungroup, and merge — with a Photoshop-style right-click menu on layer rows. Locks round-trip through PSD protected flags.

Miro integration: Pull board items into a session as references and push finished assets back to the board — no more exporting and re-uploading by hand. An admin authorizes the connector once for the workspace, then each project points at a board. See connecting Miro to Layer.

Custom fonts in video: Render timeline text and subtitles with a FONT file from your library. Opening a font now shows a type specimen — every face, a size waterfall, and a glyph grid — instead of a flat preview.

What's new?

Decompose complex imagery — multi-layer PNG and PSD export: You can now split a finished image into independently editable layers and export them as a multi-layer PNG or PSD. Buildings, props, and background come apart as labeled cutouts you can hide, refine, or composite without re-generating the scene.

Wan 3.0: Alibaba's Wan video line now produces clips of up to 30 seconds at 480p, 720p, or 1080p with optional native audio, plus first- and last-frame control so a single generation can carry a full shot rather than a silent few seconds.

Meshy V7: Image-to-3D — including multi-view from up to four angles — now produces PBR-ready meshes with game-ready topology, optional auto-rigging, and animation presets.

MiniMax Music 3: Generate complete songs up to five minutes from a music description and lyrics, and shape the arrangement with structure tags such as verse, chorus, bridge, and outro.

LTX Video 2.5: Lightricks' LTX 2.5 Pro and Fast are now available for text-to-video and image-to-video, including audio-to-video variants.

More image and video models: Grok Imagine Image 2.0, HiDream O1, Krea 2 Large, Ideogram V4, MAI Image 2.5, Wan 2.7 Pro, and Vidu Q3 Pro are now in the catalog.

New models

Seedance 2.5: ByteDance's Seedance video line takes a big step forward. You can now generate clips of up to 30 seconds with synchronized native audio — including lip-synced speech — so a single generation can carry a full spoken beat rather than a silent few seconds. Seedance 2.5 supports both text-to-video and image-to-video with first- and last-frame control for tighter shot composition, renders at 480p or 720p across cinematic aspect ratios (16:9, 9:16, 1:1, 4:3, and 21:9), and lets you dial in any duration from 4 to 30 seconds. It's built for advertising, social, and short-form production, where longer takes and on-screen speech are what make a clip usable.

FLUX.3 Video Extend and Grok Imagine Video Extend: You can now extend the duration of a generated video using FLUX.3 Extend directly from the video lightbox, with Grok Imagine Video Extend as a second extension option — giving you two distinct paths for lengthening clips without re-generating from scratch.

What else is new?

Draw-video: You can now sketch motion arrows directly onto a still image to generate an on-device reference clip, which feeds into video generation slots as a motion reference — giving you direct control over how elements in a scene should move.

Hide assets instead of delete: Assets on your board can now be hidden rather than permanently deleted, so you can keep your workspace tidy without losing work you might want to revisit.

Thinking and web search on Gemini and Luma models: Thinking level controls and web search are now available when using Gemini and Luma models in the app, giving you more control over reasoning depth and the ability to ground generations in current information.

@Reference chips now support video and audio: When referencing assets in the composer using @mentions, video and audio files are now supported alongside images — so any asset type in your library can be brought into a generation as a reference.

Drag and drop into composer: You can now drag assets directly into the composer input rather than uploading or selecting from a picker, making it faster to bring existing assets into a new generation.

Home quick actions: Edit tools are now accessible directly under the home composer, so common actions are one step closer without navigating into a specific session or asset view.