Controlled test · October 25, 2026

Which AI models keep a listing photo honest?

For staging, Nano Banana 2 and Seedream 5 Pro kept the room; FLUX Kontext Max redrew the ceiling. For video, FLUX 3 Max, Seedance 2.5 and MiniMax H3 Max kept the room; Kling 3.0 Omni added a stool and Gemini Omni Flash invented a window wall. Every model got the same photo and the same prompt, once. A listing has to show the home as it is, so the test is simple: did anything that is not furniture change?

The short answer

Stage with Nano Banana 2. Animate with FLUX 3 Max.

Best for stagingNano Banana 2

Furniture added, architecture untouched, 24 credits at 2K.

Best value videoFLUX 3 Max

Kept the room at 1080p for 87 credits per 6 seconds.

Most faithful videoSeedance 2.5

The steadiest move, at 3.7 times the price and 720p.

1,043 credits ($10.43) for the whole test: 69 for four staged images, 846 for five videos, 128 for the one-take tour; the empty room cost 24 to generate.

Part 1 · Staging

Every image model got this empty room.

Nano Banana 2 generated the empty living room as a stand-in for a real listing photo: concrete fireplace wall, exposed beams, a glass wall over a canyon. Each staging model received it as the source image with the same prompt.

Read the staging prompt
Virtually stage this empty living room for a real estate listing. Add only movable furniture and decor: a long low cream bouclé sofa facing the fireplace, a round walnut coffee table, a large textured ivory wool rug, two cognac leather lounge chairs angled toward the window wall, a tall olive tree in a matte ceramic pot in the corner, a slim brass floor lamp, a few art books and a stoneware vase on the table. Keep every architectural element exactly as it is: the board-formed concrete wall and fireplace, ceiling beams, ceiling height, window wall and its frames, the floor, the canyon view, the light direction, the camera position and lens. Do not change, move, add or remove any wall, window, door, beam or fixture. Photorealistic, same exposure and color as the original photo.
Empty living room with white oak floors, a board-formed concrete wall with a linear fireplace, exposed ceiling beams and a floor-to-ceiling glass wall over a green canyon
Source photo · empty room · 16:9

Unedited outputs

Four staging results.

Nano Banana 2 staging result

Kept the room

Nano Banana 2

24 credits · 2752×1536 (2K)

Added exactly the furniture asked for and left the walls, beams, fireplace, window wall and view as photographed. Our default for staging.

Seedream 5 Pro staging result

Kept the room

Seedream 5 Pro

18 credits · 2752×1536 (2K)

Kept the architecture and the view; chose its own furniture shapes (a curved sofa, a tall arc lamp). A good second option.

GPT Image 2.5 Sunburst staging result

Kept the room

GPT Image 2.5 Sunburst

11 credits · 1678×937

Kept the walls and windows but added things nobody asked for: framed art on the back wall, a throw, and flames in the fireplace. Smaller output.

FLUX Kontext Max staging result

Changed the room

FLUX Kontext Max

16 credits · 1392×752

Moved the camera, redrew the ceiling with different beams and added a console table. The room no longer matches the listing.

Part 2 · Video

Five video models, one staged photo, one camera move.

Same

The Nano Banana 2 staged photo as the literal first frame, the same camera-only prompt, 6 seconds, 16:9.

Nearest native setting

1080p where the model offers it at a sane price; Seedance 2.5 and Kling 3.0 Omni ran at 720p.

Pass or fail

We watched every clip frame by frame for anything that is not furniture changing: walls, windows, beams, the view, new objects.

Read the video prompt
Real estate listing video shot. Camera-only motion: a slow, perfectly smooth gimbal dolly forward into the living room toward the glass window wall, with gentle parallax between the sofa, the lounge chairs and the canyon view. The room, furniture, walls, fireplace, beams, window frames and view stay exactly as in the frame; nothing appears, disappears or changes shape. Warm late-afternoon sunlight, leaves outside sway slightly. Audio: quiet room tone only, no music, no voices.

Unedited outputs

Watch the five results.

01

Best value: kept the room

FLUX 3 Max

Cost
87 credits · $0.87
Output
6 s · 1920×1088
Kept the room
Yes
What worked

A gentle push with real parallax that held the walls, furniture and view, at 1080p for the lowest price in the test. We used it for the living room in the listing reel.

Limitation

Delivered 1920×1088, so it needs an 8-pixel crop for exact 1080p. The move is subtle.

video/flux-3-max · fal FLUX 3 Max API

02

Most faithful, most expensive

Seedance 2.5

Cost
321 credits · $3.21
Output
6 s · 1284×716 (720p)
Kept the room
Yes
What worked

The steadiest, most even push of the test; every object stayed in place to the last frame.

Limitation

Ran at 720p and cost 3.7 times FLUX 3 Max; a 1080p take costs more again.

video/seedance-2.5 · ByteDance Seedance 2.5 announcement

03

Kept the room

MiniMax H3 Max

Cost
192 credits · $1.92
Output
6.6 s · 1928×1076
Kept the room
Yes
What worked

Held the architecture and furniture through a slightly wider move toward the window wall.

Limitation

Returned 6.6 seconds at 1928×1076 instead of 6 s at 1920×1080, and cost 2.2 times FLUX 3 Max.

video/minimax-h3-max · MiniMax H3 model card

04

Drifted into a pan

Kling 3.0 Omni

Cost
108 credits · $1.08
Output
6 s · 1284×716 (720p)
Kept the room
No
What worked

Smooth motion and good light through the first half.

Limitation

Turned the push into a pan toward the back wall and added a small stool that is not in the photo.

video/kling-3.0-omni · Kling user guide

05

Invented a window wall

Gemini Omni Flash

Cost
138 credits · $1.38
Output
6 s · 1920×1080
Kept the room
No
What worked

A confident, cinematic push with natural light and room tone. On the exterior, kitchen, bedroom and terrace shots of the same listing it kept the rooms.

Limitation

In this room it pushed far past the frame and replaced the white back wall with a second glass wall within 1.5 seconds: a change a buyer would notice.

video/gemini-omni-flash · Google video generation documentation

Bonus · one take through four rooms

Can one generation tour the whole house?

FLUX 3 Max accepts up to 10 timed keyframes. We gave it the staged living room, kitchen, bedroom and terrace at 0, 5, 10 and 15 seconds: 128 credits at 720p. It cut from room to room instead of inventing hallways, kept each room as photographed, and reached the terrace through the glass door. For a listing reel, separate 1080p clips cut to music are sharper.

FLUX 3 Max · 4 keyframes · 15 s · 720p · 128 credits

Measured comparison

Cost and result, side by side.

ModelJobOutputCreditsKept the room
Nano Banana 2Staging2752×1536 (2K)24Yes
Seedream 5 ProStaging2752×1536 (2K)18Yes
GPT Image 2.5 SunburstStaging1678×93711Yes
FLUX Kontext MaxStaging1392×75216No
FLUX 3 Max6 s video6 s · 1920×108887Yes
Seedance 2.56 s video6 s · 1284×716 (720p)321Yes
MiniMax H3 Max6 s video6.6 s · 1928×1076192Yes
Kling 3.0 Omni6 s video6 s · 1284×716 (720p)108No
Gemini Omni Flash6 s video6 s · 1920×1080138No

Creative Claw sells 1,000 credits for $10. These are the credits charged in this run, not guarantees for future model versions or settings.

What this test does not prove

One room is evidence, not a ranking for every listing.

Try it on your listing

Stage, animate, label, in one chat.

Add Creative Claw to ChatGPT or Claude, attach your listing photos, and ask for an estimate before anything renders. New accounts get 100 free credits.