Signal without direction
Trend lists show what is moving, but they do not explain which promise, frame, or visual hierarchy fits your next upload.
Trends become clicks,
in one creative pipeline.
TUMB listens to the signal, lets the creator and Kimi choose the direction, then edits the supplied image into a thumbnail route ready for review.
youtube.com/feed/trending
The signal window shows what is moving; TUMB turns that movement into a compact decision brief before any image edit starts.
4 stepsfrom intent to edit
trend -> brief -> image edit · Kimi decision route · GPT-image-2 img-to-img
The problem
The hard part starts before the image edit.
TUMB keeps the signal, creative promise, user intent and source image in the same decision path.
breakpointwhy it matters
Ideas split across tools
A thumbnail starts in chat, moves to notes, jumps into image generation, then gets rebuilt in an editor. Context leaks at every handoff.
Flat output is hard to revise
A one-shot render usually ignores the source image, the creator's intent, and the reason the frame should win.
read trend and audience context
let creator intent and Kimi choose the route
edit the supplied image with GPT-image-2
return variants with a clear reason to choose
Guided discovery
From rough message to approved edit route.
The user says what they want. Kimi chooses the strongest direction. TUMB starts image editing only after the route is clear.
I have a video about rebuilding a channel from zero. Make it feel risky, honest and readable on mobile.
tumb decision rail
01 / intent
The creator describes the video.
TUMB asks for the audience, footage, target emotion and any source assets before a render is allowed.
02 / directions
Three creative routes appear.
Each route explains the promise, thumbnail composition, copy rule and why that angle should earn attention.
03 / decision
The user chooses, or TUMB decides.
The system can pick the strongest direction when the creator wants speed, but the confirmation gate stays visible.
04 / production
Generation starts only after confirmation.
The brief, title hooks and frame plan travel together into the queue so output stays attached to the original reason.
recommended route
Variant B: proof-first edit.
Promise: the proof changed everything
Frame: face on left, analytics wall on right
Copy: DAY 7, two-word proof tag
Risk: avoid tiny labels below mobile size
confirm direction
The pipeline
Six decisions, one image-edit path.
TUMB keeps every stage small enough to review: signal, intent, Kimi route, image edit, variants, recommendation.
stagedecisionoutputwhat it controls
1trend
What is moving?
signal readout
velocity, repeated hooks, niche pressure
2intent
What does the creator want?
input brief
audience, footage, assets, emotion
3kimi
Which route should win?
creative direction
promise, risk, frame rule, text rule
4edit
How should the source image change?
img-to-img prompt
composition, contrast, object emphasis
5variants
Which frames are worth testing?
thumbnail options
clean differences, not cloned renders
6ship
What should the user publish?
recommendation
best frame, best hook, why it wins
No generation starts until the decision route is approved.
route > edit > recommend
Asset-guided edit
Upload an image. Let Kimi choose the edit.
The user brings the source image and intent. Kimi turns that into a clear route. GPT-image-2 performs the img-to-img thumbnail edit so the output stays tied to the original material.
rights confirmation before upload
Kimi chooses the edit direction with the user
GPT-image-2 applies img-to-img thumbnail edits
TUMB returns review-ready variants and a recommendation
img-to-img edit rail
source image
face, product, screenshot or old thumbnail
creator intent
what should feel stronger, clearer or riskier
Kimi route
promise, composition and mobile readability decision
01
upload
02
decide
03
img-to-img
04
review
sourceuser asset
editedGPT-image-2
Kimi recommendation
Keep the original proof, increase contrast, simplify the text area, and return two stronger thumbnail edits for review.
Built for production
One edit system for three shipping modes.
Same core loop: describe the video, choose the route, edit the image, review the variants.
solo creators
01turn a rough video idea into a thumbnail/title pack before editing ends
1 approved direction, 4 frames, 8 hooks
channel teams
02keep producers, designers and editors aligned on the same promise
brief, layer notes, export handoff
agencies
03repeat a creative system across multiple niches without cloning the same look
niche atlas, variants, review history
The atlas
Niche rules, visible at a glance.
packs
13
languages
TR / EN / ES / JP
workflow
intent -> edit
Survival sprint
Cinematic hook
Reaction frame
Editorial calm
Vista hook
Day-in-life split
Habit stack
Studio illustration
gaming uses a different hook rule, but the same decision path.
mystery uses a different hook rule, but the same decision path.
food uses a different hook rule, but the same decision path.
Pricing preview
Simple plans for creators, studios and agencies.
TUMB stays focused on the core production loop: research, decide, generate, edit, ship. Limits can tighten later from real usage.
Creator
For solo channels testing titles and thumbnail directions.
$19/mo
Studio chat
Trend scan
Title packs
Thumbnail variants
Studio
For teams shipping more videos with shared project context.
$49/mo
Everything in Creator
Layer maps
Translation
Team workspace
Agency
For multi-channel operators that need repeatable creative systems.
Custom
Multi-brand context
Priority jobs
Review workflows
Custom limits
FAQ
A few clear answers before users enter the workspace.
The page explains enough of the system to make the product feel concrete without turning the pitch into an engineering spec.
What does TUMB actually do?+
TUMB turns YouTube trend signals into a structured creative brief, then ships titles, thumbnail variants and editable layered packs from a single studio. You describe the video, TUMB writes the route, generates the frame, and hands you something you can still edit.
Do I need a YouTube API key to start?+
No. You can use the studio with sample trend data and your own briefs. Connecting live data later unlocks trend ingestion for your channel and competitors while keeping the workflow simple.
Can I edit a generated thumbnail or is it final?+
Every render lands as PSD-grade layers — subject cutout, headline lockup, accent shapes, background. Swap the face, retype the headline, switch the accent color. Nothing is locked into one flat image.
How does translation keep the layout intact?+
The Translate module replaces only the text layers and re-typesets headlines so the new language fits the same visual rhythm. Subject placement, palette and grid stay identical, so the TR / EN / ES / JP versions read as one A/B pack instead of four redesigns.
What does it cost?+
Three plans — Creator, Studio and Agency. Pricing lives one section above this FAQ, and the limits will be tuned as real usage data comes in.