AI candle product photos: glass jar, flame and wax on Sume
Photograph a candle without a studio: edit one packshot with reference images on Sume, pick 1:1 or 4:5, and compare prices from $0.025 to $0.10 per image.

To get candle photos from one packshot, send the packshot as an input_references URL to an edit-capable model on Sume and describe the scene in the prompt. Qwen Image and Grok Imagine cost $0.025 per image, Flux 2 Pro $0.0375, Seedream 4.5 $0.05 and Nano Banana 2 $0.10 at its default size, all read from the catalog on 2026-10-03. Candles are a hard subject because they mix glass, wax and a small bright flame, so generate a few variants and keep the best.
Why candles are tricky
A candle jar has a curved glass wall, a printed label, a matte wax top and, in some shots, a flame. Each one is a different surface for a model to invent. The way to protect your product is to start from a real packshot and tell the model to keep the jar, label and lid identical, rather than ask for a candle from text alone. Text-only generation would redraw your label.
Pick a model and ratio
Every row below accepts reference images. n is the number of images per call; Grok Imagine returns one per call and the others up to four. Product catalog pages mostly want squares, and social feeds like 4:5, which all rows except Grok Imagine list (read 2026-10-03).
| Model | Price per image | Edits with image_urls | 4:5 | n per call |
|---|---|---|---|---|
Qwen Image (qwen/qwen-image) | $0.025 | yes | Yes | 1 to 4 |
Grok Imagine (x-ai/grok-image) | $0.025 | yes | No | 1 |
| Flux 2 Pro | $0.0375 | yes | Yes | 1 to 4 |
| Seedream 4.5 | $0.05 | yes | Yes | 1 to 4 |
| Nano Banana 2 | $0.10 | yes | Yes | 1 to 4 |
A scene prompt that protects the label
Name what must not change and give the scene one light source. Ask for a small steady flame only if you need it; unlit shots are safer because a lit flame is the part most likely to look off. Do not ask the model to render a long label text; OpenAI's image guide (read 2026-10-03) says its models can still struggle with precise text placement and clarity, so keep real label text from your own packshot.
curl -s https://api.sume.com/v1/images \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "black-forest-labs/flux.2-pro", "aspect_ratio": "4:5", "n": 4, "input_references": [{"type": "image_url", "image_url": {"url": "https://example.com/candle-packshot.png"}}], "prompt": "Keep the candle jar, label and lid exactly as in the reference. Place it on a linen-covered table beside a window, soft morning light from the left, shallow depth of field, unlit."}' What a set costs
Four variants of one scene on Flux 2 Pro are 4 x $0.0375 = $0.15. Eight scenes of four variants each on Qwen Image come to 32 x $0.025 = $0.80. Failed generations are not billed, per the docs, so retries after a failure cost nothing extra.
What Sume does not do
It does not know your real candle. Check the shape, label and color against the packshot before you publish, and keep your original photo as the source of truth.
Sources
Related posts
More in Use cases
- AI caricature from a photo via API: reference edit on Sume
Turn your own photo into a caricature with a reference edit: which Sume rows take a photo, what a call costs from $0.025, and why n=4 helps. Python example.
- AI certificate background via API: text-free art, names added in code
Generate a landscape certificate border at 4:3 or 3:2, with no text, then print names and dates with code so nothing is misspelled. Sume rows and cost.
- AI character series on Shorts: avoid the same situation each time
YouTube's inauthentic content policy flags characters in identical situations with the same outcomes. How to keep an AI character and vary the story in Sume.
- AI children's book illustrations with the same character on every page
Make storybook pages where the hero stays recognisable: draw one anchor image, pass it as a reference on every page call, and keep the style words fixed.
Written by Sume