New P-Video model
Added P-Video as a low-cost draft video option with support for text-to-video, optional start image input, optional end frame when a start image is present, 1-10 second durations, and 720p or 1080p output.
This update adds the new P-Video and Pixverse models, introduces Headshot Studio, expands start and end frame support across video generation, fixes several frame-selection and aspect-ratio bugs, and refreshes parts of the UI including notification controls.
Highlights
Pixverse and Headshot Studio
Added the new Pixverse video model and a dedicated Headshot Studio flow for generating polished square professional portraits from one or two reference photos.
Start and end frame workflow improved
Video generation now gives you more control over how a shot begins and ends, with model-aware support for flows that allow no images, only a start frame, or a guided start-plus-end-frame setup.
UI and bug fixes
Fixed frame replacement from AI Drive, cleaned up linked start/end frame slots, improved uploaded-image aspect ratio behavior, and updated notification settings with a clearer bell-based toggle.
This release focuses on video generation controls, new model coverage, image workflow expansion, and interface cleanup.
New: P-Video
We added P-Video as a new low-cost draft video model for fast iteration.
It includes:
- text-to-video
- optional start image
- optional end frame when a start image is present
- 1 to 10 second durations
- 720p and 1080p resolution support
- pricing support wired through the catalog and frontend cost display
This gives teams a cheaper motion option for rough cuts, concept testing, and quick drafts before moving to more expensive models.
New: Pixverse
We also added Pixverse as a new video generation option in the model lineup.
That gives teams another motion model to test when they want a different output character, faster experimentation, or a better fit for specific creative briefs.
New: Headshot Studio
We added Headshot Studio as a focused workflow for generating polished professional portraits.
It includes:
- up to two reference images
- selectable background styles
- selectable outfit direction
- a fixed 1:1 portrait output
- a cleaner, dedicated flow for business-ready headshots
This gives teams a faster way to create profile photos, founder portraits, team-page imagery, and other clean professional headshots without building the setup manually inside a general image tool.
Video model input handling is now more accurate
The video creation page now respects model-specific media rules instead of treating all models the same.
That includes:
- models that do not allow image input
- models that allow only a start image
- models that support start and end frames
- flows where an end frame requires a start image
This should reduce invalid UI states and prevent the page from sending unsupported inputs for the selected model.
Better start and end frame UX
We improved the frame workflow on the video generation page so you have more control over the shape of the final video.
By choosing a start frame and, where supported, an end frame, you can guide the model more deliberately:
- set a clearer opening composition or subject pose
- steer where the shot should land by the final moment
- create more intentional motion between the first and last frame
- reduce the randomness that comes from leaving the whole sequence to the model
That makes the tool more useful when you want the video to follow a specific visual direction instead of generating a loose interpretation.
The updated workflow also makes those controls easier to use in practice:
- uploaded start and end frames now appear as linked frame slots
- if you remove the end frame, the start frame remains visible with an empty end slot beside it
- if you remove both, the frame strip disappears cleanly
- once a start or end frame is selected, you can now replace it directly
- replacing a start frame from AI Drive now correctly updates the selected frame instead of failing silently
Start and end frame demo
A short walkthrough of the new start and end frame flow added in this release.
Aspect ratio fixes for uploaded images
We fixed several issues around image-driven aspect ratios:
- uploading a start image now locks the aspect ratio selector
- the selected ratio now defaults to the uploaded image ratio, or the closest supported ratio for that model
- common near-standard ratios now snap to clean values like 16:9 instead of odd fractions such as 43:24
This makes image-to-video workflows much more predictable, especially when users upload real exported assets that are close to standard formats but not mathematically exact.
UI updates
We also shipped a small interface polish pass:
- the notification settings toggle now uses a bell-based control for clearer on/off state
- video generation controls now present start and end media in a more direct, replaceable format
Bug fixes included in this branch
- fixed start frame replacement from AI Drive
- fixed frame-slot behavior after deleting end frames
- fixed cases where hidden or stale frame inputs could leak into unsupported model requests
- fixed uploaded-image aspect ratio display for standard widescreen assets
This update is mainly about making the video workflow cheaper, clearer, and harder to break while tightening up a few UI details across the product.