Disclaimer: This is an independent creative test made with AI, for demonstration only. It isn't an ad for any brand shown or named, and isn't affiliated with, sponsored or endorsed by them. All trademarks belong to their owners
Nano Banana Pro + Kling 3.0 - why my mammoth clips drifted off the frames
A woolly mammoth hands over a sling bag, and the fix for clips that ignored my frames
Tested by Nouamane Benbrahim
In today's "animal frames" series I did a spot for Mammut. A woman in a red shell jacket stands in a white studio, a larger-than-life woolly mammoth walks in and hands her a black sling bag with its trunk, and they leave before the logo hits the wall. Mammut isn't a client, this was just a test. Their logo is a mammoth, so that part was easy. This post is about the part that wasn't: two clips that totally ignored my frames, and why.
Here's my exact process.
1. The white room
I wrote the studio from a prompt alone, window on the left and shadows on the floor. Then a second pass turned the cement floor white.
Exact prompt
Create a new photorealistic image. This is a photograph taken inside a minimalist white studio from the start — not a composite. Shoot the whole space natively, with a single unified natural light source across every surface. THE VIBE: Austere, hushed, expensive emptiness — the pared-back quiet-luxury minimalism of a high-fashion campaign set before anything is placed in it. The room feels intellectual and cool, precise but not sterile, where the quality of light and the fall of a shadow are the only events. Stillness and generous negative space. THE SETTING: A clean, modern white studio — a seamless matte-white cyclorama sweep curving gently from floor to wall with no visible seam, its surface carrying the faintest real texture and tonal variation so it reads as painted plaster, not a flat digital void. A polished pale microcement floor with a soft satin sheen, subtle and almost imperceptible reflections. One tall floor-to-ceiling window just out of frame to the left. Architectural, uncluttered, refined proportions. THE LIGHT: Soft directional daylight raking in low from the left through the unseen window, casting long, clean, elongated geometric shadows across the white sweep and floor — the crisp diagonal of a window frame and mullions stretched across the wall. Gentle falloff from bright highlight to soft grey shadow, smooth gradients, no harsh clipping. Cool-neutral white balance with the faintest warmth where the sun lands. THE KEY DETAILS: The natural window-cast shadows are the subject — long, soft-edged diagonal bars and a rectangle of light on the floor. Delicate dust motes suspended in the light shaft. Subtle tonal gradation across the seamless wall. THE MOOD: Editorial, muted, cold-clinical calm with a breath of warmth. Restrained luxury. A clean neutral grade, whites held just off pure white, soft grey shadows, nothing oversaturated. THE COMPOSITION: Wide architectural frame, eye level, straight-on and balanced. Vast negative space with the shadows placed on a rule-of-thirds diagonal as the leading lines; low emphasis, open airy upper frame. THE CAMERA: Cinematic editorial architectural still — wide framing, eye-level straight position, deep focus with everything sharp edge to edge, lit exclusively by the single window daylight source with no studio strobes and no artificial separation, a clean cool-neutral naturalistic grade in the vein of Roger Deakins' understated natural light. Light falls off gently toward the frame edges. Hyperreal, tactile, every surface present.
Exact prompt
Create a new photorealistic image, but the floor is now white. Preserve everything else from the reference exactly — the seamless matte-white cyclorama wall, the soft low directional daylight from the left, the long clean geometric window-cast shadows, the composition, framing, grade and mood all stay identical. THE FLOOR (CHANGED): The floor is now a clean matte white, continuous with the seamless white cyclorama sweep so wall and floor read as one unbroken white surface, no visible seam. A soft satin sheen with subtle, almost imperceptible reflections, the faintest real texture and tonal variation so it reads as painted plaster rather than a flat digital void — never pure clipped white. THE SETTING (LOCKED): The minimalist white studio, the seamless curving cyclorama, the tall floor-to-ceiling window just out of frame to the left. Do NOT redesign the environment. THE LIGHT (LOCKED): Soft directional daylight raking in low from the left, casting long, clean, elongated geometric window-frame and mullion shadows across the white sweep and now the white floor, with gentle falloff to soft grey and smooth gradients. Cool-neutral white balance with the faintest warmth where the sun lands. Delicate dust motes in the light shaft. THE MOOD (LOCKED): Editorial, muted, cold-clinical calm with a breath of warmth, restrained luxury, clean neutral grade, whites held just off pure white. THE COMPOSITION (LOCKED): Wide architectural frame, eye level, straight-on, vast negative space, the shadows on a rule-of-thirds diagonal as leading lines, open airy upper frame. THE CAMERA (LOCKED): Cinematic editorial architectural still, deep focus everything sharp edge to edge, lit exclusively by the single window daylight source, clean cool-neutral naturalistic grade, gentle edge falloff. Hyperreal and tactile.
2. Red and cobalt
I pulled the colors from Mammut's current catalogue with their hex codes and mixed them: a vivid red hardshell (#E2231A) over a fitted base layer, cobalt-blue technical trousers (#1E62C6), a safety-orange zip pull and chunky mountaineering sneakers. Here's the first pose, made from the empty room plus text.
Exact prompt
Cinematic editorial fashion still, 16:9. A full-body figure at portrait scale occupying the LEFT third of the frame, standing a comfortable distance from camera with her whole body and some floor around her visible — generous white negative space filling the rest of the frame above and to the right. NOT a tight close crop, NOT filling the full height — a relaxed full-length editorial portrait with breathing room. THE SETTING (UNCHANGED — preserve exactly from the reference studio): the same seamless matte-white cyclorama, white curved sweep into continuous matte-white floor, faint painted-plaster texture, soft satin floor sheen. Preserve the SAME soft directional daylight raking in low from the LEFT through a window, casting the SAME long clean geometric window-frame and mullion shadows across the floor to the RIGHT — the window light and its geometric floor shadows must be present exactly as in the reference, do not flatten to plain white. Cool-neutral white balance, faint warmth, dust motes. Roger Deakins naturalistic grade, deep focus. A woman stands at full portrait scale in the LEFT third of the frame, with space around her. Late twenties, athletic, warm tan skin, dark hair in a practical braided ponytail, clean fresh minimal makeup. She wears a technical MAMMUT gorpcore look that MIXES colors between top and bottom: a vivid red (hex #E2231A) technical hardshell jacket, worn over a fitted base layer, paired with contrasting cobalt-blue (hex #1E62C6) technical trousers, and a small safety-orange (hex #F26A1B) zip-pull accent, chunky technical mountaineering sneakers. A bold mixed technical color look — red shell top, cobalt-blue bottoms, orange accent — sporty-sleek and editorial against the white studio. Confident grounded stance, weight on one leg, one hand resting at her hip, gaze off to the LEFT toward the light. High-fashion technical outdoor editorial, no large brand text.
3. The first delivery frame
My first delivery frame used that first pose. A huge woolly mammoth with shaggy reddish-brown fur, a high domed head and curved ivory tusks, holding a black Mammut sling bag in the curled tip of its trunk. She takes the bag and strokes the trunk.
Exact prompt
Cinematic editorial fashion still, 16:9, medium shot. Both figures at portrait scale — the woman the SAME size as in the first reference frame, a full-body figure with breathing room, NOT small or distant and NOT a tight crop. Generous white negative space in the upper area. THE SETTING (UNCHANGED — preserve exactly from the reference studio): the same seamless matte-white cyclorama, white curved sweep into continuous matte-white floor, faint painted-plaster texture, soft satin floor sheen. Preserve the SAME soft directional daylight raking in low from the LEFT through a window, casting the SAME long clean geometric window-frame and mullion shadows across the floor to the RIGHT — the window light and its geometric floor shadows must be present exactly as in the reference, do not flatten to plain white. Cool-neutral white balance, dust motes. Roger Deakins naturalistic grade, deep focus. The same woman from the first reference — late twenties, athletic, warm tan skin, braided dark ponytail — stands at full portrait scale on the LEFT wearing the mixed MAMMUT look: red (#E2231A) hardshell jacket, cobalt-blue (#1E62C6) technical trousers, orange accent. On the RIGHT of the frame, a magnificent WOOLLY MAMMOTH stands — a huge prehistoric elephant covered in long shaggy reddish-brown fur, with a high domed head, small ears, a long curling trunk, and massive curved ivory tusks, imposing and majestic, facing her. The mammoth has just delivered a MAMMUT technical sling crossbody bag in black with a red accent, held and offered up by the tip of its long trunk, which the woman takes with her LEFT hand. Her RIGHT hand reaches out to gently stroke the mammoth's shaggy trunk. Serene, majestic, prehistoric-meets-modern moment. Both figures clearly at portrait scale. High-fashion technical editorial, the red-and-cobalt outfit and the shaggy brown woolly mammoth against the white studio.
4. Push her back
Then I changed her distance. Over a few passes I moved her a step further back into the room, then into the left third, then one more small step back. Each pass used the earlier versions as references so her face, her skin tone and the outfit stayed put. I wrote "do NOT lighten her skin" in every one of them.
Exact prompt
Create a new photorealistic image. Keep EVERYTHING the same as the most recent version — the woman stays in the LEFT THIRD of the frame with the right two-thirds open empty white studio — but move her SLIGHTLY further back into the room, so she appears just a little smaller / a bit deeper in the space. Only a small distance change; do not recenter her, do not zoom. Preserve the EXACT same woman from the first reference — her face, warm light-brown / tan skin tone (do NOT lighten her skin), braided dark ponytail, build, and mixed Mammut outfit. Cinematic editorial fashion still, 16:9. The woman stands in the LEFT third of the frame, now a step further back in the room, full-length with floor visible beneath her and open white space above; the right side remains wide-open empty white studio. THE SETTING (UNCHANGED): the same seamless matte-white cyclorama, white sweep into continuous matte-white floor, faint painted-plaster texture, soft satin sheen. Soft directional daylight raking in low from the LEFT through a window, casting the SAME long clean geometric window-frame and mullion shadows across the floor to the RIGHT — keep the window light and floor shadows, do not flatten to plain white. Cool-neutral white balance, dust motes. Roger Deakins naturalistic grade, deep focus. THE SUBJECT: the same woman — warm light-brown tan skin, braided dark ponytail — in the mixed MAMMUT look: vivid red (#E2231A) hardshell jacket, cobalt-blue (#1E62C6) trousers, safety-orange (#F26A1B) accent, technical sneakers. Confident grounded stance, weight on one leg, one hand at her hip, gaze to the LEFT. High-fashion technical outdoor editorial, no large brand text.
5. Rebuild the delivery
Because the pose changed, I rebuilt the delivery frame at the new, pulled-back distance, with the new pose and the empty room as references. Same mammoth, same bag in the trunk, same pet on the trunk.
Exact prompt
Create a new photorealistic image. Preserve the EXACT same woman from the first reference — her face, warm light-brown / tan skin tone (do NOT lighten her skin), braided dark ponytail, build, and mixed Mammut outfit — at the SAME pulled-back distance and scale as the first reference frame (she is a full-length figure a step back in the room, NOT zoomed in). She stays in the LEFT third; the animal occupies the right. Cinematic editorial fashion still, 16:9, medium-wide shot matching the first reference's distance. Both figures full-length at that same comfortable distance, with generous white negative space above. THE SETTING (UNCHANGED): the same seamless matte-white cyclorama, white sweep into continuous matte-white floor, faint painted-plaster texture, soft satin sheen. Soft directional daylight raking in low from the LEFT through a window, casting the SAME long clean geometric window-frame and mullion shadows across the floor to the RIGHT — keep the window light and floor shadows, do not flatten to plain white. Cool-neutral white balance, dust motes. Roger Deakins naturalistic grade, deep focus. THE SUBJECT: the same woman — warm light-brown tan skin, braided dark ponytail — in the mixed MAMMUT look: vivid red (#E2231A) hardshell jacket, cobalt-blue (#1E62C6) trousers, safety-orange (#F26A1B) accent, technical sneakers. She stands on the LEFT (or crouches slightly) to receive the bag. THE SECOND SUBJECT: on the RIGHT of the frame, a magnificent WOOLLY MAMMOTH — long shaggy reddish-brown fur, high domed head, small ears, long curling trunk, massive curved ivory tusks — stands facing her. The mammoth has just delivered a black MAMMUT technical sling crossbody bag with a red accent, held and offered up by the curled tip of its long trunk, which the woman takes with her LEFT hand. Her RIGHT hand reaches out to gently stroke the mammoth's shaggy trunk. Serene, majestic moment. Both figures at the matched pulled-back scale. High-fashion technical editorial, the red-and-cobalt outfit and the shaggy brown mammoth against the white studio.
6. The logo card
Empty room, Mammut's black circle with the mammoth inside, the wordmark under it. Logo file for the artwork, the first pose for the room and light.
Exact prompt
Minimalist title card, 16:9. THE SETTING (UNCHANGED — preserve exactly from the first reference studio): reproduce the same seamless matte-white cyclorama, white curved sweep into continuous matte-white floor, faint painted-plaster texture and satin floor sheen. Preserve the SAME soft daylight raking from the LEFT through a window, casting the SAME long clean geometric window-frame and mullion shadows across the floor to the RIGHT — the window light and its geometric floor shadows must be present exactly as in the reference, do not remove them or flatten to plain white. Cool-neutral white balance, dust motes. An empty studio, no person or animal. Composite the exact MAMMUT logo from the second (logo) reference — the black circular emblem with the mammoth silhouette inside, above the black "MAMMUT" wordmark box, reproduced precisely, unchanged in shape and proportion — centered on the white wall as a flat crisp black title graphic, perfectly level and sharp. Nothing else in the frame — the empty studio exactly matching the reference, with the centered black MAMMUT mammoth logo. Editorial end-card.
7. She walks in
Empty room as the first frame, the pulled-back pose as the last. She enters from the right and stops a step back in the room, one woman only.
Exact prompt
Static locked camera, no camera movement. ONE single woman only — exactly one person in the frame at all times, no clones or duplicates. The empty white studio; the woman enters FROM THE RIGHT edge of the frame, walks across to the left, and settles into a grounded confident pose in the left third of the frame, standing a comfortable distance back in the room — weight on one leg, one hand resting at her hip, head turning to a composed gaze to the left. Natural unhurried single-person walk, the red shell jacket and blue trousers vivid against the white, long window shadows steady on the floor. Sleek technical outdoor editorial. The frame is locked; no camera movement.
8. Where it broke
New frames meant I had to redo the delivery and walk-off clips. This one burned me. My first re-renders came back with no frames attached at all. I'd passed the frames as reference images instead of binding them as the start and end frames, and on a video that gets dropped. So the model rebuilt the whole scene from my text and drifted completely off the frames I'd locked.
The fix was to bind the pose as the start frame and the delivery frame as the end frame. Then I kept the prompt short: hold the woman, her skin, the room and her spot in the left third exactly as in the frames, and only move the mammoth. It walks in from the right with the bag in its trunk, she takes it and pets the trunk.
Exact prompt
Clean white photo studio, tall window on the left, soft daylight, long mullion shadows on the floor. Hold the woman, her warm light-brown skin, red jacket and cobalt trousers, the room, scale and her position in the left third exactly as in the frames — do not redesign. Motion only: a woolly mammoth walks in from the RIGHT toward the woman, carrying a black sling bag in the tip of its trunk; it reaches her and extends its trunk to hand over the bag; she takes it and gently pets the mammoth's trunk. She stays planted in the left third — only the mammoth moves in from the right. No sliding, no backward motion, no camera push.
9. Walk-off, bound properly
Same fix here, delivery frame bound as the start, logo card as the end. I wrote the actions in numbered order: a last pet, both turn right and walk off together, then the logo snaps on in the empty room.
Exact prompt
Clean white photo studio, tall window on the left, soft daylight, mullion floor shadows. Start on the woman (warm light-brown skin, red Mammut shell, cobalt trousers) holding the black sling bag beside the woolly mammoth, as in the first frame. Action, in order: (1) the woman gives the mammoth a last pet; (2) she and the mammoth both turn to face RIGHT and walk off together toward the right edge of frame, same direction, exiting the shot — no backward or sideways sliding; (3) the room is now empty and the Mammut end-card logo (black mammoth silhouette in a circle, "MAMMUT" beneath it) snaps on all at once against the white wall, locked and sharp. Hold the window light and floor shadows throughout. No camera push.
10. Join
Emero combined the three clips into one video. Any editor does it too.
Exact prompt
Combine the clips in order
Why it works
The whole spot depends on each clip starting and ending on a still I approved. When the frames were bound, the clips held the woman, the room and the scale. When they weren't, the model was guessing from text.
Your turn: before you render a clip, check the start and end frames are actually attached as frames, not as references.
Emero writes the prompts and remembers what you've already approved, at cost.
Get one of these every day
I test new workflows daily and send them with the exact prompts, the models and the costs
5,976 generations tested so far
More workflows
See all
Nano Banana Pro + Kling 3.0 - four framings to get a flamingo hand-off right
A flamingo brings a hot-pink bucket hat, and the four framings it took to land the shot
Nano Banana Pro, Kling 3.0 · 9 min render
Nano Banana Pro + Kling 3.0 - telling a black cougar which side of the frame to sit on
A black cougar brings a woman in white her handbag, and the left and right fix that set up the shot
Nano Banana Pro, Kling 3.0 · 9 min render
Nano Banana Pro + Kling 3.0 - a peacock carries a bag by its gold chain
A gown in peacock colours, and a peacock that brings a jewel-toned bag by its gold chain
Nano Banana Pro, Kling 3.0 · 8 min render