Live creator effects
Streaming or social applications that need prompt-controlled costumes, characters, props, environments, or VFX while recording.
Independent tool overview
Lucy 2.5 at a glance
Lucy 2.5 is Decart's current real-time video-editing model for changing people, objects, attributes, backgrounds, styles, and visual effects while a camera or video stream is running. Decart says it produces higher-quality, more physically aware and temporally consistent edits live at 30 fps. Developers can use a WebRTC streaming API at 720p for $0.02 per active generation second, while asynchronous 720p edits cost $0.04 per generated second. It is a distinctive option for live commerce, streaming, advertising, social and interactive experiences, but teams must account for latency, visual errors, consent, likeness and product rights, disclosure, moderation, API-key security, and Decart terms that grant broad training rights over inputs and outputs.
Visit the official Lucy 2.5 site ↗
- Developer model ID
- lucy-2.5
- Primary mode
- Targeted real-time video editing over WebRTC
- Current resolution
- 720p in the published pricing and model pages
- Vendor performance claim
- Live generation at 30 fps with lower latency and stronger temporal consistency
- Inputs
- Video stream, required edit prompt, and optional reference image
- Core edits
- Characters, objects, attributes, backgrounds, styles, removal, and VFX
- Realtime price
- $0.02 per second of active 720p generation
- Batch price
- $0.04 per generated second for 720p Lucy 2.5 video editing
- Billing
- Pay as you go with test credits for new accounts and no listed subscription minimum
- Official SDKs
- JavaScript, Python, Swift, and Android
- API audience
- Business and professional use under the current API terms; account holder must be at least 18
- Data-use warning
- Current terms allow inputs and outputs to be used for model and product improvement, including training
- Confidentiality warning
- API terms say inputs and outputs are not confidential and prohibit sensitive data
- Reviewed
- August 31, 2026 from Decart's current product, platform, docs, pricing, terms, privacy, AUP, and DPA pages
Overview
What Lucy 2.5 is
Lucy turns an incoming video stream into a prompt-controlled live canvas. A developer can connect a WebRTC stream, provide an edit prompt and optional reference image, then change the prompt while the stream continues. The current platform lists the model identifier lucy-2.5.
The model performs targeted transformations such as replacing or adding a character or object, changing clothing or another attribute, removing objects, altering the background, applying a style, or generating visual effects. Its purpose is editing an observed scene, not generating a complete video from a blank prompt.
Decart introduced Lucy 2.5 in July 2026 with claims of better edit quality, stronger temporal consistency, more physically aware VFX, lower latency, and live generation at 30 fps. Those are vendor claims; real performance depends on network conditions, device capture, region, prompt complexity, subject motion, SDK integration, and provider capacity.
The current public API documentation prices the 720p real-time model at $0.02 per second of active generation. The same Lucy 2.5 identifier is listed for asynchronous 720p video editing at $0.04 per generated second. New accounts receive test credits, and enterprise volume terms are available.
Lucy sits inside a broader Decart family: Lucy Virtual Try-On handles garments, Lucy Restyle 2 applies full-scene styles, and Lucy Image 2 handles still-image edits. Do not attribute those separate models' features to Lucy 2.5 unless the integration explicitly selects them.
Decart's current platform and API terms deserve close review. The company says users retain input and output ownership or receive its output rights, but the terms also grant Decart broad rights to use inputs and outputs for operating, improving, training, developing and, on the consumer platform, marketing its services. API terms also say input and output are not confidential and tell developers not to submit sensitive data.
A live edit can look plausible while misrepresenting a person, product, location, event, garment fit, or physical effect. Production applications need consent, rights verification, disclosure, provenance, strong content moderation, a safe unedited fallback, and accountable human oversight.
Use cases
Who Lucy 2.5 is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Interactive advertising
Brand experiences that adapt a scene or product placement in real time with consent and clear disclosure.
Live commerce
Retail experiences that place products or environments into a live feed, while avoiding claims about actual fit or physical performance.
Gaming and virtual worlds
Interactive experiences that transform camera input or game-adjacent visuals with low perceived delay.
Event activations
Moderated booths, displays, or installations where participants knowingly opt into live transformations.
Developer prototypes
Teams evaluating real-time generative editing before committing to a larger production and compliance program.
Capabilities
Core Lucy 2.5 features
Real-time targeted editing
Changes requested elements of a running video stream while attempting to preserve the rest of the scene.
Live prompt updates
Allows an application to replace the active prompt without disconnecting the stream.
Reference-image guidance
Accepts an optional image alongside the prompt to specify a character, product, item, or visual target.
Character transformations
Changes an on-camera subject into another described character or appearance, subject to identity and artifact drift.
Object editing
Adds, replaces, removes, or changes attributes of objects in the scene.
Background and style edits
Transforms the surrounding environment or applies a new visual treatment while video continues.
Live visual effects
Generates effects intended to respond more plausibly to motion and scene geometry.
WebRTC streaming
Uses a persistent real-time connection for input and transformed output video.
Batch editing
The current pricing table also lists Lucy 2.5 for asynchronous pre-recorded video transformations.
Cross-platform SDKs
Official libraries cover JavaScript, Python, Swift, and Android integration paths.
Client tokens
Supports short-lived client authorization so a production app does not expose a long-lived API key in browser or mobile code.
Prompt enhancement
The platform example exposes an enhancement option for expanding a concise edit instruction.
Process
How the Lucy 2.5 workflow works
Step 1
Define the permitted experience
Document the allowed edit types, audience, age boundary, disclosure, capture and retention behavior, moderation rules, and prohibited high-risk uses.
Step 2
Secure rights and consent
Obtain informed permission for every person's camera, voice, likeness, biometric implications, reference image, product, brand, character, artwork, and location.
Step 3
Review the data terms
Treat consumer and API terms separately; assess Decart's training and reuse licenses, non-confidential-input clause, DPA, subprocessors, residency, deletion, security, and marketing rights before sending content.
Step 4
Keep sensitive data out
Do not stream health, financial, identity-document, children's, confidential, private-home, workplace-secret, or other sensitive footage through a default integration.
Step 5
Design a narrow prompt schema
Translate user controls into bounded edits, approved reference assets, safe attributes, and explicit exclusions instead of accepting unrestricted text directly.
Step 6
Use server-issued client tokens
Keep the Decart API key on a trusted server, issue short-lived scoped tokens, rotate credentials, rate-limit sessions, and monitor unusual use.
Step 7
Build a visible consent state
Show when camera capture and AI transformation are active, what is being changed, whether content is stored, and how to stop immediately.
Step 8
Moderate before and after generation
Screen prompts, reference images, users, and output for impersonation, sexual abuse, harassment, hateful content, illegal material, dangerous deception, and rights violations.
Step 9
Test realistic failure modes
Measure end-to-end latency, jitter, dropped frames, occlusion, multiple people, fast movement, lighting changes, network loss, prompt changes, unsafe prompts, and service interruption.
Step 10
Provide a safe fallback
On timeout, quality failure, policy uncertainty, or disconnect, stop the transformation and switch to an approved blank, paused, or clearly unedited state.
Step 11
Validate representation
For products or garments, verify color, logo, dimensions, placement, coverage, fit, material, and claims; do not present a synthetic preview as proof of real-world performance.
Step 12
Label synthetic edits
Preserve watermarks or provenance and add clear, persistent disclosure when a reasonable viewer could mistake the edited stream for reality.
Step 13
Control cost and abuse
Set session-duration, concurrency and credit limits, alert on spikes, stop inactive generation, and test billing under reconnect and retry behavior.
Step 14
Require human accountability
Keep a moderator or operator able to stop live output; review recorded or commercial content before publication, targeting, purchase decisions, or reuse.
Cost
Lucy 2.5 pricing and free plan
Decart lists pay-as-you-go pricing with no subscription and no minimum spend. Lucy 2.5 costs $0.02 per second of active 720p real-time generation and $0.04 per generated second for asynchronous 720p video editing. New accounts receive free test credits and enterprise volume pricing is custom. Credit expiration language differs across current consumer-platform and API terms, so verify the exact purchase surface or order form before prepaying a large balance.
Test credits
Free initial credits
For evaluating model quality and integration before funding sustained usage.
- Available to new accounts according to the current pricing FAQ
- Amount is not published on the public documentation
- Credit validity and permitted uses should be checked in the account
- Not a permanent free tier
Lucy 2.5 Realtime
$0.02 per active second
For 720p live editing sessions over the real-time API.
- Billed per second of active generation
- 30 seconds costs $0.60 at the published rate
- 60 seconds costs $1.20 at the published rate
- Pausing or disconnecting inactive sessions promptly matters for cost
- Network, storage, moderation, streaming and application costs are separate
Lucy 2.5 Batch Video
$0.04 per generated second
For asynchronous 720p edits of pre-recorded video.
- Billed per generated output second
- 5 seconds costs $0.20 at the published rate
- 30 seconds costs $1.20 at the published rate
- Retries and alternate outputs increase total spend
- Input storage, transfer and downstream rendering are separate
Enterprise
Custom
For higher volume, negotiated commercial terms, security review, or support.
- Contact Decart for volume pricing
- Request an order form, service levels, DPA, security materials, subprocessor list, and incident terms
- Negotiate input and output use, model training, confidentiality, retention, deletion, indemnity, and support where necessary
Pricing checked . Check current pricing at the source ↗
Assessment
Lucy 2.5 strengths and limitations
Where it stands out
- Purpose-built for live editing rather than only offline generation
- Current Lucy 2.5 endpoint is documented and active
- Targeted prompts can modify part of a scene while trying to preserve the rest
- Prompts can change during a persistent WebRTC session
- Optional reference images provide more concrete visual guidance
- Published 720p pricing is simple and inexpensive per second for short tests
- Same model family supports both real-time and asynchronous editing
- Official JavaScript, Python, Swift, and Android SDK paths
- Client-token architecture supports safer browser and mobile integration
- Separate Restyle, Virtual Try-On and Image models let developers choose a narrower purpose
What to consider
- Vendor claims of 30 fps, lower latency and stronger consistency are not guarantees for a particular application or network
- 720p is the only resolution currently listed for Lucy 2.5 in public pricing
- A live edit can flicker, drift, lag, freeze, hallucinate, occlude the wrong object, or change elements that should remain fixed
- Fast motion, hands, faces, hair, transparent objects, reflections, multiple people, low light and scene cuts are demanding
- The transformed stream can misrepresent real people, products, garments, places, events, safety conditions or physical effects
- Lucy 2.5 is not the same as Lucy Virtual Try-On, Lucy Restyle or Lucy Image; those require separate model selections
- Reference images and camera streams can contain biometric, personal, location, household, child, confidential or copyrighted data
- Current platform and API terms grant Decart broad rights to use inputs and outputs for improvement, development and model training
- Consumer platform terms also describe content use for marketing or promotional purposes
- API terms say input and output are not confidential and instruct developers not to submit sensitive data
- The DPA does not automatically make every sensitive or regulated use appropriate
- Developers must provide lawful notices and consent and implement their own content moderation and safety controls
- Current API terms prohibit impersonating real people and require deepfake or AI disclosure where applicable
- Output may not be unique and may be similar to another user's result
- Commercial output rights do not clear copyright, trademark, publicity, privacy, character, music, product or endorsement rights
- WebRTC connectivity, camera permissions, firewalls, mobile constraints, region and server availability affect reliability
- A long-lived API key must never be embedded in client code; short-lived tokens and a trusted server are required
- Per-second billing can grow quickly across concurrent always-on sessions, reconnect loops, abuse or forgotten streams
- Consumer-platform and API credit-expiration terms differ, so large prepaid purchases need contract-level confirmation
- Live deployment requires an immediate kill switch, safe fallback, disclosure, logging, incident response and human moderation
Compare
Lucy 2.5 alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Content Creator
Runway Aleph
Choose Runway Aleph for managed in-context edits and transformations of existing footage in an offline creative workflow.
Explore Runway Aleph →Content Creator
Kling O1
Choose Kling O1 for multimodal hosted video generation and editing through text, image, and video inputs.
Explore Kling O1 →Content Creator
LTX-2 Model Family
Choose the LTX-2 family when you want open weights, self-hosting, synchronized audio-video generation, and current API options.
Explore LTX-2 Model Family →Design
Adobe Firefly
Choose Adobe Firefly for a broader creative suite and closer integration with established editing, design, rights and enterprise workflows.
Explore Adobe Firefly →Content Creator
Pika
Choose Pika for a simpler creator-oriented video-effects and generation interface rather than a custom real-time API product.
Explore Pika →Content Creator
Higgsfield AI
Choose Higgsfield for a broader hosted social-video creation platform with templates, camera motion, image and video models.
Explore Higgsfield AI →Questions
Lucy 2.5 FAQs
What is Lucy 2.5?
Lucy 2.5 is Decart's current real-time world and video-editing model. It transforms selected people, objects, attributes, backgrounds, styles or effects while an input video stream continues.
Is Lucy 2.5 a text-to-video model?
Not in the usual blank-canvas sense. Its primary input is a live or recorded video plus an edit prompt and optionally a reference image. It transforms the observed scene.
How much does Lucy 2.5 cost?
The current public pricing is $0.02 per second of active 720p real-time generation and $0.04 per generated second for asynchronous 720p video editing. New accounts receive an unspecified amount of test credit.
What resolution does Lucy 2.5 support?
Decart's current public model and pricing pages list 720p for Lucy 2.5. Confirm the SDK model constraints and account before building around another resolution.
Does Lucy 2.5 run at 30 fps?
Decart says the 2.5 release generates live at 30 fps. Treat that as a vendor claim and benchmark end-to-end frame rate, delay, jitter, stability and quality in the exact region, network, device and scene conditions you will deploy.
What can Lucy 2.5 edit?
Decart describes character swaps, added or removed objects, product replacement, attribute changes, backgrounds, style changes and visual effects. Results and preservation of untouched areas are not guaranteed.
Is Lucy 2.5 the virtual try-on model?
No. Decart has a separate Lucy Virtual Try-On model family for garment-specific workflows. Lucy 2.5 is the general targeted video editor.
Can I use Lucy 2.5 in a web or mobile app?
Yes. Decart provides WebRTC integration and official JavaScript, Python, Swift and Android SDKs. Keep API keys server-side and issue short-lived client tokens to untrusted clients.
Does Decart train on Lucy inputs and outputs?
Current platform and API terms grant Decart rights to use inputs and outputs to operate, improve and develop services, explicitly including AI-model training. The consumer platform terms also mention marketing use. Organizations should negotiate and document any required exception before sensitive or proprietary use.
Can I send confidential or sensitive video to the Lucy API?
The current API terms say input and output are not confidential and instruct developers not to submit sensitive data, including health information. They also reference a DPA for personal-data processing. Do not assume that a DPA overrides the product restrictions or training license; obtain written terms for the exact use.
Who owns Lucy 2.5 output?
Decart's current terms say the user or developer retains input and output rights, or receives Decart's output rights, subject to compliance. The terms also grant Decart a broad license, and output may not be unique. Third-party rights remain the user's responsibility.
How should live AI edits be disclosed?
Use a clear persistent label when viewers could mistake the transformed feed for reality, preserve any watermark or provenance, disclose material product simulations, and follow applicable deepfake, advertising, platform and AI-transparency law.
What happens if the live model fails?
A production app should fail to an approved paused, blank or plainly unedited state; never leave a frozen or misleading synthetic frame looking live. Provide an immediate kill switch, timeout, health checks and a human moderator.
Does Lucy 2.5 guarantee virtual product accuracy?
No. A synthetic product placement can get geometry, color, logo, material, fit, scale, shadows or occlusion wrong. It must not be presented as evidence of actual fit, safety, availability or product performance.
Bottom line
Our Lucy 2.5 verdict
Lucy 2.5 occupies a genuinely differentiated niche: prompt-controlled edits that happen while video is live, at a published price low enough to prototype interactive commerce, streaming and event ideas. The API surface is clearer than the original landing page, and the current $0.02-per-second live rate is easy to model. The hard part is governance, not the demo. Broad contractual training rights, non-confidential inputs, facial and product misrepresentation, per-second abuse, deepfake disclosure and live failure handling make Lucy suitable only for consented, moderated, tightly scoped experiences with a safe fallback.
Visit Lucy 2.5 website ↗Similar Content Creator Tools
View all Content Creator AI tools →
Runway Dev
Runway Dev- New developer API for Runway's media models and real-time avatars
beehiiv Agent
Beehiiv Copilot - Beehiiv's built-in agent for newsletter analytics, segments, and campaigns
Seed Audio 1.0
Seed Audio 1.0 - ByteDance's model for generating speech, music, and SFX in one pass
FLUX 3
FLUX 3 - BFL’s new multimodal visual intelligence model with 20-second generations, now in early access
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.