Character IP Short‑form Project
Nyangbanjip is a short-form channel following the everyday lives of Oat and Kongi, two cats with distinct personalities. Since July 24, 2026, I have developed the characters, produced the episodes, and run the account. Alongside Seedance, I now use MiniMax H3 and publish videos generated locally in ComfyUI.
- Year
- 2026
- Team
- Individual
- Role
- Solo character and episode development, image and video generation, local ComfyUI workflow setup, editing, publishing, and audience analysis.
- Tools
- Topaz Video AI, Premiere Pro, After Effects, Photoshop, Seedance 2.0 / 2.5, MiniMax H3, ComfyUI, ChatGPT Image
Followers
TOP 3
Most-viewed Reels
- TOP 01Views5.66M
- Reach
- 3.92M
- Likes
- 655.7K
- Shares
- 152K
- Saves
- 28.1K
- TOP 02Views4.8M
- Reach
- 2.82M
- Likes
- 387.1K
- Shares
- 78.8K
- Saves
- 13.6K
- TOP 03Views1.76M
- Reach
- 1.05M
- Likes
- 113.9K
- Shares
- 32.5K
- Saves
- 5.8K
Challenge
The project needed to preserve both characters' appearance and personality across generative production while keeping a repeatable format from feeling like the same story each time.
My part
- Defined two recurring characters through distinct appearance, personality, relationship, and behavior rules
- Maintained identity with at least three character references, adding scene-specific object and product-placement sources alongside prompt, shot, and QC criteria
- Generated films with Seedance and MiniMax H3, configured local ComfyUI workflows, and handled upscaling, editing, sound, and captions
- Compared views, reach, and follower conversion, then fed the findings into the next episode
Same characters, a new story every time
Directing each character’s reaction in ComfyUI
In ComfyUI, I connect separate references for Oat, Kongi, and their relative scale to MiniMax H3. I define the action and camera in the prompt and connect a prepared soundtrack to guide the scene’s timing.
I designed “Want a Bath?” around two contrasting reactions: Kongi darts away after the owner’s question, while Oat takes its time. Each cat starts out busy with something else, and the gap in their reaction timing shows the difference in personality.
Lock appearance, personality, and relationship
Defined appearance, scale, expressions, relationship, and recurring behavior before producing episodes.
Minimum three references
Added scene-specific object and product-placement references to the character references, using them in the prompts to integrate characters and props into the same scene.
Change the event, keep the rhythm
Kept a Hook–Action–Result–Reaction rhythm while changing each episode's cause and ending.
Generate and select shot by shot
Generated and compared shots with Seedance and MiniMax H3, checking appearance, grounded paws, both characters' paths, and reaction timing before selecting a take.
- ChatGPT Image
- Seedance
- MiniMax H3
- ComfyUI
Upscaling, editing, sound, and captions
Checked each shot's color and audio continuity and applied grading and upscaling where needed, then finished the edit, sound, and captions for publishing.
- Topaz Video AI
- Premiere Pro
- After Effects
Feed response into the next episode
Compared views, reach, and follower conversion, then applied the strongest situations and pacing to the next brief.
Result
The most-viewed episode passed 5 million views, and the channel is still running.
I manage character references alongside scene-specific objects and product-placement sources, then feed views, reach, and follower conversion back into the next episode's situation and editing pace.
Credits
Character direction, generative production, editing, and publishing · Lee Joon-suh