Clip-Ahead
Generates the next complete clip before the current clip finishes playing.
- Output
- 5–15 second clips
- Primary metric
- Real-Time Factor
METHODOLOGY
Evidence before adjectives.
Video Runtime separates what a provider says from what we have measured. This page defines the rules behind every published status, performance claim, price, and verification label.
01 · DEFINITION
What counts as real-time?
“Real-time AI video” covers different systems. We classify a project by how visual output is produced and how quickly a user can affect it, not by a marketing label alone.
Frame-Streaming
Generates a continuous visual world frame by frame while a user or agent controls it.
- Output
- Continuous frames
- Primary metric
- Frames per second and interaction latency
Live Transformation
Transforms a live camera or video feed while it is being captured or streamed.
- Output
- Live video input
- Primary metric
- End-to-end latency and frames per second
02 · MEASUREMENT
Real-Time Factor, with context
RTF compares the time a user waits for a playable result with the duration of that result. Lower is faster, but it is not the only measure that matters.
Real-Time FactorEnd-to-End Generation Time ÷ Output Duration
Potential to prepare the next clip while the current one plays. This is not a guarantee of seamless playback.
Near real-time, but likely to need buffering or a transition.
Not suitable for a seamless Clip-Ahead experience.
Backend inference is not end-to-end time
A provider’s inference time can omit upload, queue, orchestration, decoding, and delivery. We only call a result end-to-end when it measures from submission to playable output. Region, queue, input, resolution, P95 latency, and failure rate can all change the experience.
03 · HANDS-ON STANDARD
How a hands-on result earns its label
A single fast result is not a test. We require repeatable timing, a documented environment, and retained evidence before publishing a hands-on claim.
- Use the same device, browser, and region for the test set.
- Complete at least three full runs and report the median.
- Record submission time, playable time, and output duration.
- Record login, queueing, failures, test date, and configuration.
- Retain a rights-cleared screenshot or recording as evidence.
- Show an approximate P95 only with at least five runs, and never use the fastest run as the representative value.
04 · STATUS
Access status has a fixed meaning
- Live
- Publicly accessible, with the primary experience working.
- Beta
- Accessible, but explicitly described as a test or beta.
- Developer-only
- An API, repository, or developer endpoint is available, but there is no public consumer experience.
- Waitlist
- Only a waitlist or invitation request is available.
- Restricted access
- An age, password or other access gate is visible. This does not prove the service is offline or that playback works.
- Paused
- The official page remains, but the primary experience is paused.
- Offline
- The primary experience is inaccessible without a pause notice.
- Unknown
- Current access cannot be confirmed; it stays out of the main index.
05 · VERIFICATION
Evidence labels stay separate
Hands-on tested
We completed the documented test procedure and retained dated evidence. Results identify the environment and number of runs.
Official source checked
The claim comes from a linked first-party page, documentation page, endpoint page, or official repository. It is not presented as our measurement.
Unverified draft
Evidence is incomplete or not current enough to publish. These records remain private and are never indexed.
06 · PRICE EVIDENCE
Prices are dated, not assumed permanent
Every selectable rate keeps the model, endpoint, resolution, billing unit, source URL, verification date, and effective period. Standard rates remain separate from temporary launch offers. Expired prices stay historical, and prices are checked again before launch or after a material provider change.
See source-linked cost estimates →07 · EDITORIAL INDEPENDENCE
Commercial relationships do not buy conclusions
Affiliate links never change a verdict. Sponsored work must be labelled. A project team may submit a correction, but it cannot purchase a positive finding. We disclose relationships on the relevant entry and publish the evidence behind material claims.
08 · UPDATES
When the index changes
V1 does not promise a fixed recheck schedule. Every entry shows its last verification date; older evidence is not silently treated as current, and launch decisions require a fresh check.
Before publication
A record stays private until its required sources, status, verification label, and update date pass review.
After a material change
Access, model, price, license, performance, or source changes trigger a new check and a dated update.
When evidence is uncertain
We remove or downgrade the claim instead of filling a gap with an estimate. Unverified drafts are never public.
09 · RANKING RULES
How projects earn their position
Each list answers a different question. None is a universal quality score.
Recently added
Projects appear after editorial approval, in their reviewed categories, ordered by first publication in this index. Later approvals appear first. Edits, withdrawal and republication preserve the original time. Duplicate submissions are reviewed against the existing entry. A project moves down as newer projects are added; this is not a penalty.
GitHub stars and growth
Total stars use official repository snapshots checked within seven days. The separate growth list subtracts a same-repository snapshot seven days earlier from a snapshot checked within 48 hours. We allow up to one extra hour of collection drift and show both observation dates. Equal counts or gains share a rank. Zero and losses are valid; missing, stale or insufficient history is excluded and explained.
Editorial discovery
Important updates require a reviewed, source-linked change to availability, price or capability. Routine edits cannot refresh a position. Editor’s picks include a reason and do not follow traffic, payment or submission time. We have no public popularity score.
These GitHub lists cover reviewed repositories in this index, not all of GitHub. Snapshot dates identify actual collection times; this version does not promise automatic daily refreshes. A missing observation is never reported as falling popularity.
Explore the lists →