Okaeri Darling!: a finished AI anime music video and the takes that failed

The maker provides the full video and says the song preceded it.

Jill Mama embeds the complete Okaeri Darling! MV in her production note. The song had already been released through Spotify and other services; this project gave form to her imagined welcome-home scenes and dog-eared idols on stage. Drawing on her music-industry experience, she conceived the scenes herself rather than asking AI to invent the whole story. Watch how the room, entrance and stage alternate and how screen time is shared across three singers.

The film is playable, but its audience and earnings are not established here. The maker hoped a video might help the song spread. She also says the song had already appeared in TikTok uses before the MV; that cannot be counted as a result of this video.

• Watch the complete MV on the maker's channel ↗

Sources for this section: Jill Mama: Okaeri Darling! animated MV production note · The complete Okaeri Darling! MV

The maker limited cast and sets before mapping lines and shots.

The maker initially designed five Italian-greyhound-inspired idols, then reduced the cast to silver-, black- and brown-haired characters. She found that hair, clothes, positions and faces were easy to mix across a larger cast; three also made individual singing moments easier to allocate. Names and personalities helped manage many assets and direct movement. These were the maker's planning decisions, not a structure automatically supplied by a model.

She chose three spaces—the room, the entrance and the stage—and two costume sets, then used Excel to map lyrics, singers, poses, camera ideas and generated stills. She did not complete every board before starting; clear ideas were fixed first and other shots evolved with the material. For a music-video trial, aligning lyrics, cast, sets and shot intent before long motion runs makes rework easier to diagnose.

• Read the maker's cast and Excel board account ↗

Sources for this section: Jill Mama: Okaeri Darling! animated MV production note

An unlocked 16:9 frame, microphone and detailed clothes caused more work.

The maker used ChatGPT Images 2.0 for most stills, but initially judged poses and expressions without building each image for the final 16:9 frame. Video then carried black bars; specifying 16:9 at generation did not solve it, so she remade many stills. The note shows part of the board, not every before-and-after input and output. We can identify the stage and the reported correction, not fabricate a controlled comparison.

Stage characters initially held microphones. The maker reports that motion made microphones vanish, multiply, fly off or become dark rods. Some earlier takes worked, so the final edit mixes shots with and without microphones. Frills, ribbons and other costume details added in the stills were also hard to preserve across shots; she regenerated conspicuous changes and accepted smaller accessory differences. This was a design and tolerance decision, not a proven fix from adding more adjectives to a prompt.

Sources for this section: Jill Mama: Okaeri Darling! animated MV production note

Wardrobe drift, lip-sync failure and a successful camera move come from one maker and project, not one shot before and after revision.

In the wardrobe section, the maker embeds a Seedance 2.0 failure in which the character's clothes and hair drift visibly. A second embed shows a failed attempt to sync a short phrase to the mouth. These are failed clips the maker chose to share, more concrete than a generic warning about glitches. The note does not provide complete prompts, fixed seeds, retry counts or independent corrected versions of those exact shots.

She also shares a camera move she liked: she prepared high and low visual references, specified camera movement closely and left character action relatively open. It is a different shot, not a correction of the wardrobe or lip-sync failures. Her experience was that longer prompts were not always better and brief directions could sometimes yield livelier motion. This is one project's observation, not a controlled result for every model or genre.

Sources for this section: Jill Mama: Okaeri Darling! animated MV production note · Maker's wardrobe-and-hair drift clip · Camera move the maker liked · Maker's lip-sync failure clip

The maker selected moments in CapCut, changed speed and placement and sometimes returned to stills.

The maker says a generated 15-second clip might yield only ten useful seconds because its motion or ends do not fit. A segment might be sped up to around eight seconds to meet the beat; another good-looking clip might need a different place in the song. She repeatedly edited, checked for jarring transitions, swapped material and regenerated, sometimes starting again from a still. The final video is a selection of usable moments against the music, not a row of complete model outputs.

She wanted short opening and closing lines to sync, but found sung syllables and dance beats difficult to align, accepting the level she could achieve with that workflow. The maker lists Suno as her music tool; the song was complete and released before this MV project. For the video, she lists ChatGPT Images 2.0 for stills, Seedance 2.0, Veo, Kling and Sora for motion, and CapCut for her edit. The public note does not assign every finished shot to a particular model, nor does video generation account for the whole music-making process.

Sources for this section: Jill Mama: Okaeri Darling! animated MV production note · The complete Okaeri Darling! MV · Maker's lip-sync failure clip

The maker gives an approximate subscription-and-credit spend and describes an enquiry that did not become a job.

The maker estimates around ¥4,500 for two months of Firefly and extra credits, plus around ¥10,000 for CapCut and additional Seedance use, rounding the total to about ¥15,000. The parts add to roughly ¥14,500, underscoring that this is a rounded personal account, not a ledger. This sum covers those two tool categories, not her existing paid Suno use or days of image and editing work. The page offers no complete invoices, release costs or income. It is not a reproducible quote for anyone to make an MV.

The maker says the song had been used on TikTok and that a potential client asked about AI animation for a seminar. After she explained the time and budget involved, no commission followed. These are possible paths to attention or service work, not evidence that this MV caused the song's earlier uses or secured an order. The public material includes no platform settlement, client receipt or net profit for the video.

Sources for this section: Jill Mama: Okaeri Darling! animated MV production note

Keep a cast and lyric plan plus inputs and failed outputs; test continuity and edit usability early.

One possible trial starts with a piece of music you may use, one character, one set and two shot tasks: a performance and a story beat. Make stills for the final aspect ratio, keep the character and clothes simple, then test motion separately. Record references, prompts, outputs, cost and the seconds that survive the edit. One attractive still cannot establish continuous video quality. This small trial is BitShovel's suggestion drawn from the account, not the maker's guaranteed recipe.

The maker makes the MV and process account available to view, not for others to republish her song, characters, footage or notes. A paid attachment appears below the article; BitShovel has not used or summarized it. Anyone planning a commercial MV should check rights for the song, sound, characters and each generation service. The public record does not establish a repeatable profit model for this project.

• Read Jill Mama's original production note ↗

Sources for this section: Jill Mama: Okaeri Darling! animated MV production note