To make Instagram product photos with AI that look like one feed rather than a pile of experiments, approve one "style master" image and attach it as a reference to every new post, together with photos of the product you are showing. On an 8frame board that is a master on Nano Banana 2, one generation node per post with the master and the product attached, and the original PNG downloaded for posting. The limit people do not expect: among 8frame's image generators, Nano Banana is the one with a 4:5 option, the ratio Meta recommends for feed images, so a set made on another model has to be generated at 3:4 and trimmed.
TL;DR
| Step | Node or model on 8frame | Fixed price | Input |
|---|---|---|---|
| 1. Style master | Nano Banana 2 at 4:5 | 8 / 11 / 17 / 22 credits at 0.5K / 1K / 2K / 4K | Mood references, one product |
| 2. Each post | Nano Banana 2 at 4:5 | Same per image | Master plus product photos (up to 10 references) |
| 2b. Posts with many references | Seedream 5 Lite at 3:4 | 6 credits per image | Up to 14 references; trim to 4:5 |
| 3. Fix one detail | Flux Edit or Kontext Max | 7 for a ~1 MP input (rising with size) / 13 | The post image |
| 4. Enlarge if needed | BRIA Increase Resolution | 6 credits, 2x or 4x | The approved image |
| 5. Post | Download "Original" | No credits | The PNG file |
Why AI feeds drift
A grid is judged as a whole: nine to twelve squares seen at once on a profile. Four things make an AI-made grid look patchy:
- No shared image. Each post made from text alone starts fresh, and the colour temperature, surface and light wander.
- Words that change. "Soft morning light" in one prompt and "warm natural light" in the next are two set-ups to a model.
- Mixed models. Two models given the same prompt and reference make two looks.
- Mixed ratios. One post at 1:1, the next at 4:5, the next cropped from 16:9 gives a feed with different amounts of background around the product.
The fix is the same as for a catalog, applied to a feed: one master, one model, one ratio, one set of words.
The board, node by node
The closest live template is PRODUCT PLACEMENT on /workflows. This recipe adds the style master and the feed rules.
Step 1: decide the feed's four constants
Write these on a sticky note at the top of the board and paste them into every prompt:
- Palette: two or three colours that appear in every post (a surface colour, a background colour, an accent).
- Light: one direction and quality ("soft light from the left, gentle shadow to the right").
- Surface or setting family: "pale stone and linen", "dark wood and brass", "bright solid colour backdrops".
- Framing: how much of the frame the product fills and where ("product centred, about half the frame height").
Keep the sticky short. Four lines you actually reuse beat a page of mood words.
Step 2: make the style master
Attach two or three mood references (your own photos or past posts that you like) and one product, and generate on Nano Banana 2 at 4:5. Nano Banana 2 is the default Nano Banana model, priced by size: 8 / 11 / 17 / 22 credits at 0.5K / 1K / 2K / 4K. Draft at 1K. When one image looks like your feed should look, make it again at 2K or 4K and call it the master.
The master is not necessarily a post. It is the reference that every post copies its light and colour from.
Step 3: one node per post
For each post, make a generation node with the master attached first, then two or three photos of the product in the post (front, three-quarter, a detail). Nano Banana 2 takes up to 10 references, so there is room for the master, product photos and a prop. Prompt with the four constants plus one line for this post's idea ("the candle beside an open book", "the candle held in two hands").
If you need more references per image (a set of products in one shot, plus props, plus the master), Seedream 5 Lite takes up to 14 at 6 credits per image. It has no 4:5 option: generate 3:4 and trim about 6% of the height in your editor to reach 4:5. Do not mix the two models in one feed run; pick one for the set.
Lay out the post nodes on the board in a three-column grid in posting order. The board then shows you the feed before you post it, which is the point: you see a post that breaks the set while it is still a draft.
Step 4: fix details by edit, not by regeneration
When a post is right except for one thing (a prop that competes with the product, a shadow in the wrong direction), edit it. Flux Edit is 7 credits for a ~1 MP input, rising with input size; Kontext Max is 13 and also offers 4:5. One change per edit. Regenerating the whole image to fix a prop risks changing the product.
Step 5: download the original
8frame stores the image as delivered (PNG for Nano Banana 2 and Pro, Seedream 5 and GPT Image 2, the models we measured) and offers four sizes in the download menu: Small, Medium and Large, which are WebP versions 256, 640 and 1280 pixels wide, and Original, the stored file. Meta's ads guide recommends 1440 x 1800 pixels for 4:5 feed images. Large, at 1280 wide, is under that. Download Original for posting and ads.
If the original is smaller than you need, enlarge it with BRIA Increase Resolution (6 credits, 2x or 4x, content-preserving) before posting. Clarity (13 credits) is diffusion-based and always enlarges 2x on 8frame, and can redraw fine product detail; avoid it for products.
A monthly rhythm
For a feed that runs for months, keep a simple rhythm on the board: one frame per month, holding that month's posts in posting order, with the master at the top. At the start of each month, duplicate last month's frame layout, swap the post ideas, and keep the master and references unchanged. Changing the look is then a deliberate decision (a new master, dated), not something that happens by accident between two posts.
Which model for which step
| Need | Model | Price | Ratio note |
|---|---|---|---|
| Master and posts at 4:5 | Nano Banana 2 | 8 / 11 / 17 / 22 by size | Offers 4:5 |
| Premium look for one hero post | Nano Banana Pro | 21 at 1K or 2K, 42 at 4K | Nano Banana family |
| Many references per image | Seedream 5 Lite | 6 | No 4:5: make 3:4 and trim |
| Text-led posts | GPT Image 2 | 29 square, 20 other ratios at default high | Check its ratio list; set text yourself |
| One-detail fixes | Flux Edit / Kontext Max | 7 (rising with size) / 13 | Kontext offers 4:5 |
| Size for posting | BRIA | 6 | Content-preserving |
Feed images versus feed ads
For a comparison of AI product photo tools beyond 8frame, see our product photo generator roundup; this section is about the feed specs.
The same images often run as ads. Meta's ads guide for Instagram feed image ads, checked 2026-11-09, recommends 4:5 at 1440 x 1800 pixels, accepts JPG or PNG up to 30 MB, and suggests keeping primary text to 125 characters and headlines to 40. Keep the words in those fields rather than in the image: image models redraw lettering, and text set in the ad is editable and translatable.
If a post later becomes a video for Reels, see Instagram Reels ad specs for the 9:16 rules; a 4:5 still does not become a 9:16 clip without recomposing.
Where it breaks
- The product changes between posts. Colour, label and shape drift on regeneration. Attach two or three real photos every time and compare each output with the product. See why AI changes product labels.
- Master drift. If you change the master halfway through a month, the grid splits in two. Version the master: keep the old one on the board in a frame named with its dates.
- Mixed ratios. One model at 4:5 and another at 3:4 crop differently. One model per set.
- Posting the Large download. 1280 pixels wide is under Meta's recommended 1440. Download Original.
- Text in the image. Prices, offers and handles drawn by the model drift. Set them in a design tool or the caption.
- Real people. If a post shows a person, use a model you have consent from or a generated person who does not resemble anyone real, and say so where the platform asks. More in synthetic talent and likeness in ads.
FAQ
How do I keep AI product photos consistent on Instagram?
Approve one style master, attach it as a reference to every post with real photos of the product, use one model and one ratio for the set, and reuse the same four lines (palette, light, surface, framing) in every prompt.
What size should AI product photos be for Instagram feed ads?
Meta's ads guide recommends 4:5 at 1440 x 1800 pixels, JPG or PNG, up to 30 MB.
Which 8frame image model makes 4:5?
Among the image generators, Nano Banana offers 4:5; the Kontext Max edit node offers it too. Seedream 5 Lite does not: generate 3:4 and trim about 6% of the height.
Which download size should I post?
Original. Small, Medium and Large are WebP versions 256, 640 and 1280 pixels wide; Original is the file as delivered (PNG for the Nano Banana, Seedream 5 and GPT Image 2 files we measured).
How many reference images can I attach?
Nano Banana 2 takes up to 10, Seedream 5 Lite up to 14. Our table of reference image limits by model has the rest.
Sources
- Meta Ads Guide, Instagram feed image ads (4:5, 1440 x 1800 recommended; JPG or PNG; 30 MB; primary text 125 and headline 40 characters): facebook.com/business/ads-guide, checked 2026-11-09
- 8frame prices, defaults and limits (ratio options per image node, download sizes, PNG output): the 8frame app, checked 2026-11-09
Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.