Repository navigation
FreeVideo 0.5.3 - #806
Merged
Merged
FreeVideo 0.5.3#806
Conversation
…agnostic report When the retained copy of FreeVideo.app behind the desktop shortcut does not verify, the launcher saves what failed (codesign exit and message, whether the running app itself verifies, stray ._ files, the volume type) to launcher/desktop-shortcut-failure.json, which the diagnostic report includes. The user-facing message and the verification itself are unchanged; the record is removed once a copy verifies.
…itor Node tests run again
…ne, with Retry and Report below
Importing an offline package again finishes a model file whose copy stops early, in the package store and in the install. Only a shorter file whose bytes equal the start of the verified original is completed; any other differing file is never overwritten. A refused rename or interrupted copy removes the temporary file.
…package again The update status says whether the folder was installed from offline packages, and the repair advice for a model that is not fully installed names the Installation page of the launcher in that case.
The count the server returns covers every kind, so it now shows once under the characters, scenes and props rows instead of under each, and no longer calls every item a character. Also refreshes two out-of-date browser tests (Autopilot tabs and toolbar, the could-not-read lines).
…empty prompt switches
…s retry and report only... Connection card: the arrow wraps with its words, a failed lookup shows retry and report only, and the button icon keeps its gap
…Detect & review before... Explain missing or damaged model files on the failure card, and name Detect & review before Install / repair in page repair hints
…and sizes just under a unit... Settings: plan sizes under 1 GiB show MiB or KiB instead of 0.0 GiB, and sizes just under a unit boundary carry to the next unit
Failure cards, the editor's failure rows and the dubbing note whose hint says to repair in the page settings now offer a secondary Open settings button. It opens Settings through the existing opener at Detect & review, scrolled into view and focused. Launcher and offline-pack hints are unchanged and show no button. A browser test covers the failure popup and the Studio card in both languages.
…led like a lost connection The queue and history reads checked only the status before parsing JSON. A page shown while the engine restarts, or one put in front of it by a proxy or tunnel, has a success status, so parsing threw a syntax error that skipped the retry and the notice that clears itself and stopped a repeating request on the first read. Such a reply is now raised like a lost connection, with the status, content type and kind of body in the details.
The picture page reads how many references each model takes from the engine instead of fixed numbers, so H3 Image opens up once the engine passes pictures on. Over a model's limit, Generate stays solid and points to the extra pictures. The note for pictures left out now stays visible, and two-digit labels fit their cells.
Library-list cards stored an already-prefixed thumbnail address, and the library prefixed it again, so behind ComfyUI's /api prefix the thumbnails failed to load. Store the plain route like server-sent sources. Adds a browser check with a prefixing API in Chinese and English at 1440 and 400 wide.
…nd recognise the refusal in... Keep running when Windows application control refuses a SciPy file, and recognise the refusal in any system language Transformers and Diffusers treat SciPy as optional but import it unconditionally once it is installed; a refused SciPy file stopped the GPU check and every generation. Both now see SciPy as not installed when it cannot be loaded. A file Windows refuses to load is identified by its error code instead of its system-language text, and the GPU check shows the Windows protection card for it. The CosyVoice vocoder imported SciPy at module load, so a refused SciPy file stopped dubbing before any audio was made. Every audio use of SciPy now loads it when needed and reports a refusal with the Windows error code, so it shows the same card as other refused files.
…ead of failing every retry The pinned Diffusers source is built once and checked on later setups. A folder left without its Git data (removed by a cleaner or sync tool, lost objects) made the check fail with a bare Git exit code, and every retry failed the same way. Such a folder is now kept beside the new one under a separate name and built again. A readable folder with another tree is still never replaced, and Git that cannot start is still reported as it is.
Two chapters (33 -> 35): snapping clips to the beats of the music on Audio 2, and the system status in the top bar with Stop now. Chinese and English pictures captured from the shipped panels with offline fixtures.
…fit before starting, and... H3 Image takes up to 16 reference pictures, refuses ones that cannot fit before starting, and runs long sequences in row slices References are sized, labelled and encoded as the released edit pipeline does and enter the sequence as fixed conditioning rows. Long sequences run their row-wise work in slices of 16384 rows, which gives identical results while avoiding fragmented card memory. The planner counts reference rows from sliced measurements, and requests whose streamed-only floor cannot fit the card are refused with the reference-memory message before the text encoder runs.
…try once the group is two... A refused whole-sequence buffer no longer takes another head-group retry once the group is two heads or fewer
…rs that the same setup... The GPU precheck before an int8 switch no longer requires model folders that the same setup downloads afterwards The precheck before switching an installation to int8 only checks that the GPU is ready; the model folders are checked after the install, including when the earlier GPU test results are reused.
…ss, and add the public npm... Keep slow Python runtime downloads patient when a mirror refuses access, and add the public npm mirror as a source
…ge's temporary reference... Image generation: save results as high-quality JPEG and remove the page's temporary reference and mask uploads after each job
… generation, opening from... Waits say what is happening: prompt rewriting, size range, starting a generation, opening from the launcher, subtitle unpacking and Autopilot show their stage and elapsed time; a refused part keeps its reason on the generate bar
…under Save, the... A damaged reserve setting is kept aside, a refused save is explained under Save, the command-line upload shows the check stage and reference preparation counts files A damaged reserved-memory setting file is renamed and kept, and the settings panel and generation continue with automatic optimization. A failed save and a value under the lower limit are explained directly under the Save button. The command-line upload reports when the file is being checked. Preparing reference media reports which file is in progress.
The film video is unchanged; each part is noted with its shot's action, ready for finishing the sound.
A screenplay carries its outline and the blocking of each scene; the start confirmation lists the outline
…confirmation when a project... Imported clips can be removed from the editor library, with a second confirmation when a project uses them
…it of a finished short drama
…and Edit, and the Edit on... Chat: a finished short film comes as a finished-video card with Open and Edit, and the Edit on result cards is always visible The chat read lists each film that joined successfully with its play, thumbnail, size, length and work; the chat shows it as the card a finished video uses, and the newest result, a video or a film, leads. Edit opens the edit the film's shots were cut into. The hover-only style of a message's Edit no longer hides the Edit on finished-video cards. Built-in directors are given a short while to finish when the film service closes.
H3 Image: the same prompt and references are encoded once; later images reuse the saved text conditioning
…es within the card and... H3 Image: the encoder, DiT stack and VAE stay resident between pictures within the card and memory budget; the plan counts what is kept and the process's own CUDA context
… loads, and every reference... H3 Image: kept models give the card back before the reference encoder loads, and every reference memory check uses the budget less the CUDA context
…ir hint, Chinese node... Node tests follow the offline repair import, the Detect & review repair hint, Chinese node labels and the progress label fake DOM
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
0.5.3: Short films are finished for you: the built-in director picks and joins the takes, the film arrives in Autopilot as a card, and an editing project with one part per shot is ready. H3 Image takes up to 16 reference pictures, and consecutive images start sooner. Imported clips can be removed from the editor, and waits, failures, setup and narrow windows read more clearly.