Interior design mood board to video: up to 10 reference images

To turn an interior mood board into a video, send up to ten of its images as references to Gemini Omni Flash 1.1 and point to each one in the prompt with a tag like <IMAGE_REF_0>. On Sume an 8-second clip costs $1.00 at 720p and $1.50 at 1080p, billed at the vendor list price times 1.25.

The model uses the references as a guide to the furniture, colours and materials. It does not place your exact product in a fixed position, so treat the result as a mood film for a client, not as a product render.

What Omni Flash accepts as references

The Sume catalog lists these limits for reference-to-video on Gemini Omni Flash 1.1.

Build the board

• Choose three to six images, not ten. Every image the model has to honour makes it harder to follow all of them.
• Use one image per object: sofa, rug, lamp, wall colour. A photo of a whole room mixes the objects together.
• Add one or two clips of 3 seconds or less if you want a camera move, such as a slow pan, copied.
• Crop each image to the object, with a plain background where you can.

Point to each reference in the prompt

References are numbered from 0 in the order you list them. The first image is <IMAGE_REF_0>, the second <IMAGE_REF_1>, and a reference clip is <VIDEO_REF_0>. Name what you want from each: "the sofa in <IMAGE_REF_0>", not just "the sofa".

The request below uses three images and an 8-second 16:9 clip.

import os
import time

import requests

API = "https://api.sume.com/v1/videos"
HEADERS = {"Authorization": "Bearer " + os.environ["SUME_API_KEY"]}
body = {
    "model": "gemini-omni-flash-1.1",
    "prompt": "Slow walk-through of a living room using the sofa in <IMAGE_REF_0>, the rug in <IMAGE_REF_1> and the lamp in <IMAGE_REF_2>, warm evening light",
    "duration": 8,
    "resolution": "720p",
    "aspect_ratio": "16:9",
    "input_references": [
        {"type": "image_url", "image_url": {"url": "https://example.com/sofa.jpg"}},
        {"type": "image_url", "image_url": {"url": "https://example.com/rug.jpg"}},
        {"type": "image_url", "image_url": {"url": "https://example.com/lamp.jpg"}},
    ],
}
job = requests.post(API, headers=HEADERS, json=body, timeout=60)
job.raise_for_status()
poll_url = job.json()["polling_url"]
while True:
    state = requests.get(poll_url, headers=HEADERS, timeout=60).json()
    if state["status"] in ("completed", "failed", "cancelled"):
        break
    time.sleep(10)
print(state["status"], state.get("usage"), state.get("unsigned_urls"))

References or first and last frame

If you want the clip to begin on one exact photo, use image-to-video with frame_images instead. A single reference image with no first or last frame is treated as reference-to-video, so the photo guides the clip and does not open it. Use frames for a reveal of one room and references for a mood film that combines many objects.

References also change the cost of exploration. Because the price follows seconds and resolution, a board of three images costs the same as a board of ten. Spend your effort on picking images, not on counting them.

Check before you show a client

Compare each furnishing in the clip with its reference. Models often change a pattern, a leg shape or a colour. If an object is wrong, drop the other references and test that object alone at 360p, which costs $0.30 for 8 seconds. Omni Flash always produces sound, so mute the clip if you plan a voice-over.

A sensible plan for a client room is three steps. Render a 360p test at $0.30 to check that every object appears. Render a 720p version at $1.00 for the client review. Render the 1080p final at $1.50 only after sign-off. For five rooms with one 360p test, two 720p renders and one 1080p final each, the total is $19.00: five tests at $0.30 are $1.50, ten 720p renders are $10.00 and five finals are $7.50.