Fren Pet Costume Contest 2026 - Recap Video - Production Notes ============================================================== Deliverable artifacts/fren-pet-costume-contest-2026.mp4 Technical checks (ffprobe on the final file) Duration ........ 212.0 s (3:32), 6360 frames Video ........... H.264 High profile, yuv420p, 1920x1080, 30/1 fps, CRF 18 (x264 preset slow) Audio ........... NONE - no audio stream (the brief says "NO AUDIO TRACK") Faststart ....... yes, the moov atom comes before the media data (offset 36) File size ....... 50,719,827 bytes (48.4 MiB, under the 64 MiB limit), average about 1.9 Mbit/s Frames pulled back out of the encoded MP4 and checked: captions are sharp and readable, and the Results table is readable. Note on audio The brief says "NO AUDIO TRACK" and the file follows it. The generic acceptance line mentions "H.264 video with AAC audio". This was not done, because the requester's own spec rules out an audio track. The MP4 is H.264/yuv420p with faststart and plays in browsers without audio. If an audio track is needed after all, a silent AAC track can be muxed in without re-encoding the video. Scene runtime (start times; 1 s crossfades at S1->S2, S2->S3 and S5->S6, other changes are continuous) S1 Opening hook ............ 0:00.0 H fills the frame; orange title, then green subtitle; Swarm Icon 1 left, Icon 2 right S2 Meet the entrants ....... 0:10.5 A-G revealed one at a time, left to right, on an invisible upward-bowing arc; "These 7 Pets were entered by their players" -> "Now let's see who won!" S3 Winner podium ........... 0:28.0 hay-bale podium; A in 3rd -> E in 2nd -> F in 1st with confetti, glow and sparkle S4 Barn-door reveal ........ 0:42.5 podium and pets swing away in perspective; the frog walks the hanging door open across its rail; four text groups appear in order; push through the doorway; upward pan through the starry sky to T1 S5 How the judging worked .. 1:07.5 T1 fills the frame, then pulls back; T1-T5 shown with alternating layouts; orange "https://api.imd.fun" stamp mid-T2; 18 colored diamonds around T5 S6 Reports and results ..... 2:04.5 report 1-7 flip in and out in order beside the retained caption; then Results.PNG large and held about 12 s S7 Vote highlights ......... 2:44.4 downward pan through the diamond-star sky to H; four text beats; gold aura on F, moonlight-silver aura on E, bronze aura on A, each fading with its text S8 Closing invitation ...... 3:16.0 H kept on screen; greeting, invitation with @frenpetonbase, sign-off; held to 3:32 Caption text (shown exactly as specified, no step-number prefixes; no "top builds, updated weekly" footer) Vote figures shown: 18 judges / 17 votes / 1 abstention; 16 judges agreed on 1st (F); 2nd 11 votes over 6 (E); 3rd 16 to 1 (A). These match Results.PNG. Handle shown: @frenpetonbase Assets and how they were handled - Pets A-G: cut out of the supplied PNGs by removing the connected light-mint backdrop and ground band. Only background pixels were removed; pet pixels were not repainted or changed. Every costume, prop and companion is kept: A's staff, E's easel, F's sword and shadow companion, and D's wisps. Scaled with nearest-neighbour only (1.3x in the lineup, 1.5x on the podium). - OPEN-CLOSE-SCENE.png: fitted to the 1080 px height without cropping or stretching. The two 120 px side bands are a blurred, darkened extension of the artwork itself. In S7/S8 it sits as a rounded card, with nothing cropped beyond a 14 px corner radius. - Winner auras: the exact silhouettes were found by template-matching the F/E/A cutouts inside H. Glows and outlines are drawn only outside each silhouette, so no costume pixel is covered. - Screenshots, reports and Results: only framed and resized (Lanczos); their contents are unchanged. - 18 judge diamonds: 6 colors sampled from the swatches in color ref.png (red, orange, yellow, green, cyan, lavender), 3 of each, shuffled so no two neighbours share a color. Each diamond is a single solid color with no gradient. - Stars that were added are white open-outline diamonds in varied sizes. The original stars in H are untouched. - Generated elements (drawn in code, no AI image generation): twilight autumn backdrop, hay-bale podium with pumpkins and vines, confetti, barn wall, hanging sliding barn door on its rail, the frog, the meadow and the starry skies. - Typography: Montserrat (titles) and Inter (captions), both SIL OFL, from Google Fonts. Material limitations - The frog is a simple flat-vector character (green, blue shirt, tan pants, short), drawn from behind with its head lowered so its face never shows. It is a stylised stand-in rather than a detailed illustration. - The S4 swing-away is a 2.5D perspective warp of the podium group, not a true 3D camera move. During that 2.6 s move the pets are resampled bilinearly; everywhere else pet pixels are nearest-neighbour. - The pet cutouts come from soft-edged screenshots, so a faint 1 px dark edge pixel can remain on some outlines. It belongs to the source, not to the effect. - The opening title is in H's sky and passes above the crescent moon, close to it. - The T5 caption sits in the screenshot's empty upper-right area at 33 px, so it is smaller than the other captions to fit the space. - There is no audio, as specified. - Checking was done on sampled frames, not by watching the whole video in real time. Sources video-src/prep.py - pet cutout extraction and locating winners in H video-src/render.py - frame generator (scenes, timing, typography); frames encoded with ffmpeg