The AI Filmmaking Pipeline
Spot the slop
Slop hides in stills and multiplies in motion. Inspect the exact crop you plan to use, name the visible defect, then decide whether that crop is safe to pass into Seedance.
The four tells
Most slop is one of these four. Flip through the deck until your eye runs them automatically:
Light with no transitions
Flat-black pits instead of a smooth shadow ramp — and they transfer onto every character you add.
Broken-but-plausible objects
Crates, railings, hardware you can almost read — in motion they turn to mush and multiply.
Local logic breaks
An effect in only part of the frame — rain that scratches one corner. Seedance's logic breaks with it.
Oily textures
Soapy surfaces lose their material; reflections crawl in motion, so the plate cannot hold continuity.
A visible still-frame defect is already a stop. Use Seedance only when the crop passes the still scan and you need motion to confirm an uncertain edge, reflection, or object.
Two models, two slop accents
Beyond the four universal tells, every model has its own fingerprint. The two you'll lean on most for generation slop hardest in these specific ways — learn both and you can already name the culprit from a single frame.
Banana slop (Nano Banana Pro) — ruler-straight symmetry, everything parallel and set square, flat light and color with no contrast play, so the frame is pretty but staged and lifeless, like stock photography. Textures are detailed yet read as a 3D render, not a photograph. And it hyperbolizes every edit: ask for graffiti on a wall and the whole location gets tagged; ask for "more alive and cinematic" and it takes you literally, littering the frame with junk and dirt.
GPT slop (GPT Image 2) — sharpness and microcontrast cranked to the ceiling, hard halos on every edge. No depth: everything in focus, no bokeh, no separation between planes. White balance pulled warm until the frame yellows, materials go plastic and licked-smooth, and the dynamic range gets squeezed to the middle — shadows and highlights meet halfway and the frame goes limp.
The team ran the same farmhouse brief through both. Same location, two accents:
The texture pattern, up close
GPT's most damning tell is a single sickly texture laid over the entire frame — the handbook's own example is this garage, where the same film-wrap pattern sits on the cabinets and on the car bodies alike.
What clean looks like
The handbook's instruction is literal: stare at good frames until your eye calibrates. Browse these — each one holds the exact thing slop breaks.
Your turn: pick the frame apart
Two real Soul Cinema locations from the team's slop file — each hides four tells. Treat the full 2.32:1 stage as the intended shot crop and expand it when you need to inspect the source pixels. Before you tap, scan the crop and predict what each defect would break downstream; pins stay hidden until you click close enough to reveal one.
Pizza corner, take one — find the four tells
1/2
Click on the image where you think a tell is hiding.
Keep training your eye
Two frames build the habit; they won't cover every tell you'll meet. We built a dedicated Spot the Slop practice app with a deeper pool of real frames, timed rounds, and no repeats. Run it before you choose a location or approve a final shot.
Practice Spot the Slop