AI Cartoon: A 24-Second Episode Built Ten Ways, and the Best Cost 162 Credits

The team that makes the vertical cartoon series “Forest Street” entirely with cloud models has published a production breakdown. Between 29 and 31 October 2026, it rebuilt one episode, “Campaigning Is Prohibited”, in about ten ways using Wan, Kling, Veo, Seedance, GPT Image and Nano Banana Pro. The episode runs 24 seconds, with recurring characters and Russian speech. Costs are given in Higgsfield credits (the internal currency of the service where Kling runs) and in Flow bonuses (how Veo is billed). The two units cannot be added together. No method has audience numbers yet, and every trial was a single run, so the breakdown cannot be read as a model ranking.

Order of steps: a cheap draft first, animation later

The pipeline is as follows: a character bible, a second-by-second script, human approval, reference character sheets, frames in GPT Image, ElevenLabs voices, assembly with camera movement, checks, and a human review of the draft. Only then is the scene handed to Kling 4.0: one continuous shot, 2–3 takes. That way, animation money goes only to a story that has already passed review.

The authors' main conclusion: the weak spot was vague prompts, not the models themselves. The script is written out second by second: who speaks, with what delivery, what the listener does, where the pause falls, when the object is revealed. For the third Kling prompt, the team added a ban on echo, exact intervals between lines and a “ZINA ONLY” note. All five lines were recognized without repeats. Before release, a fresh AI assistant retells the scene from frames without sound. If its reading does not match the intended meaning, the episode does not ship.

Ten trials on one story

All trials date from 30 October, on the story “Campaigning Is Prohibited”, about 24 seconds long:

• Wan 3.0 Prime + Speak 2.0, shot by shot: 116 credits, about 47 minutes.

The clothespin is already in the hand, and the letters were drawn in on their own. Rejected by a human. • Kling 4.

0 + Avatars 2.0, shot by shot: 61.75 credits, about 17 minutes.

Extra letters on the T-shirt, and Zina blinks instead of squinting. Rejected. • Veo 3.

1 Fast, shot by shot: 180 bonuses, about 29 minutes. The character sheet moved instead of the clothespin. Rejected.

• Images with camera (“comic”): no new costs. Noticeable shot changes; rejected. • Kling 4.

0 in one go: 54 credits, about 6 minutes. The vatrushki (cheese pastries) are visible in advance, the board shows pseudo-text, and the room changes. Screened out by the automated check.

• Seedance 2.5 in one go: 168 credits, about 8 minutes. Two Zinas, wrong captions, no sound.

Screened out. • Veo 3.1 Fast, three parts: 80 bonuses, about 10 minutes.

Characters blended together, the vatrushki are not revealed, and loudness is −18.3 LUFS. Screened out.

• Kling 4.0 with second-by-second direction, 3 takes: 162 credits, about 16 minutes. One shot, five lines, vatrushki revealed in the finale.

Accepted. • GPT Image + camera, 10 frames: 12 generations, 373 characters of voiceover. Poses change in jumps, and the text has no errors.

Accepted as the cheap version. • Nano Banana Pro 2K + camera (unlimited): poses change instantly, and the pigeon is in the wrong place. No rating.

In a blind test on six clips (162.5 credits), the human reviewer liked Wan 3.0 Prime best. But on a scene with dialogue and emotion, Kling performed better, and the choice did not hold. Seedance was dropped from production over quality, even though an unlimited plan covered it. Video models mangle letters, so all sheets and boards in the video are now clean, and the lettering is drawn in GPT Image and Nano Banana images. Video is generated at 720p and upscaled to 1080×1920. A Kling generation at 1080p cost 96 credits versus 54.

Voice and checking: what automation cannot hear

Voiceover comes from ElevenLabs eleven_v4 with emotion tags. For the eight characters, the team ran a casting of 24 candidates (2 992 credits). Zina's first voice, Nadia, was dropped: it is tagged as young, and the character is 74. Babushka Valya was chosen, but by its description, not by ear. Slow recordings are sped up to ×1.30 without changing pitch.

Kling speaks in its own voices, so the speech is replaced. Audio Isolation separates the voice, the lines are cut into windows, each one passes through Voice Changer (eleven_multilingual_sts_v2) and is laid over a quiet background. The video stream is unchanged: its SHA-256 hash is identical before and after. Word boundaries shifted by 0.06–0.08 s at most, according to Scribe, ElevenLabs' speech recognition service. That does not confirm lip articulation.

Checking is kept separate from assembly. Frames are captured every 0.25 seconds, about 100 per clip, and an AI assistant reviews them. Speech is recognized by a local faster-whisper small. Loudness is measured with ffmpeg and normalized to −14 LUFS. The native audio track fluctuated: Kling at about −11.7, Veo at −18.3 with a dip to −23.3 in the third part. No AI assistant can hear audio: uploading a WAV returned “audio input not supported”. Recognition returns words, not stressed syllables, so it cannot check stress. A 31-word dictionary could not be attached: the API key lacked the pronunciation_dictionaries_write permission.

What remains unsolved

Carrying a good image over into video is unreliable. In “Where does the phone go?”, the box covers the tray, the lamp shifts with no visible action, and the phone rises before the pause.

Tyoma's character sheet makes him look younger than 16, and every frame inherits that. The image-based versions double silhouettes on five-frame transitions. There is no by-ear confirmation of the voices.

The main thing the authors do not know is how many rubles and human hours it takes to get a good episode regularly, rather than one lucky attempt. The practical takeaway for anyone making AI video: write the scene out second by second, test it on cheap frames, and do not trust an automated check report without your own ears.