Creator Canvas — Video

Drop an Avatar node from the toolbar to add a Theo Anchor presenter scene. Avatar nodes are siblings of Voiceover and Music — each carries its own script, avatar pick, voice pick, aspect ratio, motion prompt, expressiveness preset, and render lifecycle. Wire it into the Result node and it becomes one presenter scene in the final cut.

Script is required. The avatar reads your script verbatim. The visual prompt and reference image surfaces from the regular Scene node do not apply — use a Scene node when you want a b-roll shot instead.

Pick from the gallery. Click the presenter tile on the Avatar node to open a visual gallery. The picker lists every avatar your workspace has access to (filter chips narrow to Avatars, Talking photos, or All), and each tile carries a face preview. Each tile carries a Theo Anchor–compatible signal sourced from the upstream avatar catalog. When that signal is missing, the gallery stamps an amber “May not render” pill on the tile and raises a confirm-time toast — you can still pick it (some avatars render successfully anyway), but if Theo Anchor returns an error at submission time the Avatar node surfaces a “try another presenter” hint. Leave either control on Default to fall back to the stock presenter / voice.

Browse voices visually. The voice tile opens a full gallery dialog with search across name, language, and gender, plus chip filters for language (long-tail languages collapse behind a +N more popover), gender, and an Emotive toggle that narrows to voices that support emotion tags. Each card has a play / pause button with an animated waveform — only one preview rolls at a time, and the footer context bar tells you which voice is playing or staged for selection. Hit Use this voice to commit your pick (Cancel leaves the node untouched).

Pick a backdrop. The Backdrop tile opens a popover with curated color presets, a custom color picker, and an image uploader. Drop a brand still (or pick from connected Asset nodes) to put the presenter in front of a custom backdrop. When both color and image are set, image wins.

Chain multiple presenters. Add several Avatar nodes for a multi-host deck. Each one picks its own avatar + voice, and the timeline strip orders them alongside Scene nodes when you drag the running order.

Fallback is automatic. If Theo Anchor is unavailable at render time, the chain falls back to a Theo Reel b-roll clip rendered from the same script so the deck still ships.

Every Scene node accepts an uploaded image as a render anchor. Drop a screenshot, dashboard, illustration, or photograph onto the dropzone and the renderer uses it as the first frame, so the motion stays faithful to your upload instead of inventing an unrelated visual.

Theo Auto routes based on your content. When you leave the engine on Theo Auto and attach a reference image, Theo routes deterministically based on what kind of image you uploaded:

• Screenshots, dashboards, charts, documents, photos → Veo 3.1. Best text rendering in the catalog (Vidguru benchmark: 5/5 vs Kling 2.6 1/5).
• Illustrations, animation frames, drawings → Seedance 2.0. Cinematic motion on stylized content; Arena #2 ranked.
• Prompt mentions “draft” / “quick” / “preview” → LTX-2. Cheapest premium tier for rapid iteration.
• Prompt mentions “multi-shot” / “sequence” / “scene transition” → Seedance 2.0. Multi-shot storytelling specialist.

We deliberately do not silently fall back to a different engine on a failure here. The previous fallback path (Kling 2.6 by default for Auto + image) produced renders that ignored uploads with text on them. If you want a different look, pick a specific engine on the pill row.

Per-engine behaviour. Each engine treats the reference frame slightly differently:

• Veo 3.1 uses your image as the first frame and animates outward — highest fidelity to text and detail.
• Sora 2 anchors physically accurate motion on your image — great for natural movement and long takes.
• Seedance 2.0 drives multi-shot sequences from your image — best for illustrations, animation, and dynamic motion.
• LTX-2 renders fast, high-resolution iterations from your image — best for drafts and budget-friendly clips.
• Hailuo 02 brings people in your image to life with micro-expressions — best for portraits and character work.

The renderer matches your reference image to the scene’s aspect ratio (16:9 by default, 9:16 for vertical). When your image already matches the target within ±5% the bytes are preserved verbatim. When the aspect differs, the renderer center-crops to fit and tells you what happened: the “Anchored on your image” pill appends either “Cropped to match your aspect ratio” or “No crop needed — your aspect ratio already matched.”

Detected: … pill (beta). When the content-classifier feature is on for your workspace, Theo runs a quick vision pass on your upload and shows the detected content type (dashboard screenshot, photograph, illustration, etc.). It’s informational today — future iterations may use it to drive smarter routing.

If the anchor fails to load, the Scene node surfaces a clear “Anchor failed” pill with the reason — we never silently substitute a fabricated frame, so you always see why a render didn’t match your image.

The Avatar node is opt-in per workspace. When the underlying service isn't configured, the avatar / voice picker shows a locked hint pointing operators at the environment variable documented in the deployment guide. Setting that key on the server unlocks the node for every user without restarting the canvas.

Scene nodes (Theo Auto + every brand-name engine) and the Voiceover / Music nodes all keep working when Theo Anchor is unavailable — only the Avatar node is gated.

After a render lands, the Scene node surfaces a small badge telling you which engine actually shipped the clip:

• Theo Auto · rendered with Veo 3.1 when you left the engine on Theo Auto.
• Theo · rendered with Sora 2 when you explicitly picked the engine.
• On a rare fallback, an amber Theo · you picked Sora 2 — finished on Veo 3.1 pill shows up with a one-click Retry button so you can re-render on your original engine pick.