Lock product identity; vary only the shot and scene
One watch reference transformed into five consistent ecommerce images
Ecommerce teams often face a simple but expensive problem: the supplier provides one product photo, while the marketplace expects five hero images and the product detail page needs even more lifestyle, detail, and usage shots.
The fission-pattern skill is designed for that gap. Codex reads one reference image, locks the product's shape, color, material, hardware, and structural details, and then prepares separate generations for different camera angles and scenes. The goal is not five loose variations. It is the same recognizable product shown as a front hero, a 45-degree view, a lifestyle image, a macro detail, and a scale or interior composition.
In this tutorial, you will:
1.Install the fission-pattern skill in Codex;
2.Prepare a source image that can support a consistent set;
3.Use a dry-run to review five commands and estimated usage;
4.Inspect the set and rerun only the failed image.
Important: fission-pattern is a workflow that Codex reads, not an offline photo editor. A third-party dLazy cloud service using gpt-image-2 generates the images, so you need a dLazy account, the CLI, and available credits.
Two types of image sets
Product set
Generate a front hero, a 45-degree dimensional view, a lifestyle image, a macro detail, and a scale comparison or interior view of the same product.
Pose set
Keep the same model and outfit while generating front, side, back glance, walking, and seated poses.
The same model and outfit in five different poses
Both modes share the same boundary: do not change the product's shape, color, material, or structure, and do not invent features, promotions, or endorsements that do not exist.
Prepare the source image first
How to prepare a clean product reference
The source image should meet these requirements:
File size between 20KB and 15MB;
Resolution above 400×400;
JPG, JPEG, PNG, or WebP format;
The full product structure is visible and not covered by a hand or prop;
The background is clean, with no marketing text or watermark.
Describe the selling points precisely. “Lightweight” is broad; “100% cotton, lightweight and breathable, with French-inspired contrast detailing” gives Codex enough direction to choose a cafe setting, natural window light, and a restrained film grade.
Install the fission-pattern skill
The original skill is available here:
GitHub: dlazyai/ecommerce-skills — fission-pattern
Option A: Ask Codex to install it
Create a new Codex task and enter:
$skill-installer
Install the fission-pattern skill from this GitHub repository path:
https://github.com/dlazyai/ecommerce-skills/tree/main/skills/fission-pattern
Check the entry filename during installation. If the repository uses
lowercase skill.md, rename it to uppercase SKILL.md and verify that
Codex can discover the skill.
Restart Codex after installation. Type $, or run /skills in the CLI or IDE extension, and confirm that fission-pattern appears.
Option B: Install it manually
For project-only use:
.agents/skills/fission-pattern/SKILL.md
For user-level access across projects:
~/.agents/skills/fission-pattern/SKILL.md
Install the dLazy CLI and sign in
Run the version pinned by the skill:
npx @dlazy/cli@1.2.3 --help
Or install it globally:
npm install -g @dlazy/cli@1.2.3
Device login is recommended:
dlazy login
If you already have an API key:
dlazy auth set YOUR_API_KEY
Never include a real key in a public conversation, screenshot, or Git repository. Local images, prompts, and parameters are uploaded to the third-party cloud service when the task runs.
A practical five-shot recipe
A standard five-image product set can be organized like this:
1.Front hero: clean seamless background, even studio light, maximum clarity;
2.45-degree view: show depth, thickness, and the side profile;
3.Lifestyle image: map the selling point to a controlled use scene;
4.Macro detail: prove material, stitching, markers, hardware, or craft;
5.Scale/interior image: use a ruler, reference object, or internal view.
One detail is easy to misunderstand: do not replace five shot prompts with one --batch 5. Batch generates multiple versions of the same shot. A complete set requires five prompts with five different shot segments.
The consistency rule: change only the third segment
Workflow from source image to fidelity lock, five shots, and set review
Each prompt has three parts:
Product fidelity segment (identical across the set)
- Selling-point atmosphere segment (identical across the set)
- Shot segment (different for each image) For a watch, the fidelity segment might be: The subject is a polished stainless-steel watch with a silver sunburst dial, applied baton markers and a black crocodile-embossed leather strap. Keep the product 100% faithful: same case shape and polish, same dial colour and marker layout, same hand shapes, same crown, same strap embossing and stitching. Reuse this segment word for word in all five prompts. Only the final camera and scene description should change. That tells the model it is creating different photographs of one product rather than five similar products. First task: ask Codex for a dry run Place the source image in the project: docs/fission-pattern/product-watch.jpg Then enter: $fission-pattern
Create a five-image ecommerce set from:
docs/fission-pattern/product-watch.jpg
Product: stainless-steel watch with a silver sunburst dial.
Identity details: polished round steel case, silver sunburst dial,
applied baton markers, three hands, one crown on the right,
black crocodile-embossed leather strap, and black stitching.
Selling points: business-ready, precision craft, vintage leather character.
Prepare separate commands for:
- Front hero; 2. 45-degree view; 3. Worn at a cafe;
- Dial and crown macro; 5. Top-down scale image on dark navy blueprint paper.
Reuse the exact same product-fidelity segment in all five prompts.
Change only the shot segment. Use 1024x1536 and jpeg.
Use quality high for the macro and medium for the other images.
No text and no watermark.
Run a dry run first. Show all five commands and estimated usage.
Do not generate yet.
After approval, reply:
The commands and estimated usage look correct. Generate the set and save it as:
docs/fission-pattern/output-watch-1.jpg
through docs/fission-pattern/output-watch-5.jpg
Map selling points to controlled scenes
Warm/padded: snowy street, cold ambient light, warm rim light;
Breathable/quick-dry: gym or running track, dynamic action, bright daylight;
Waterproof: rainy pavement, beading droplets, overcast light;
Business/commute: office lobby or subway station, cool neutral light;
French/vintage: cafe interior, marble table, warm window light, film grading;
Skin-friendly/baby: soft nursery textiles, pastels, very soft diffusion;
Precision/craft: dark navy blueprint paper, brass ruler, directional side light.
If the environment overwhelms the subject, add:
The product must occupy at least 40% of the frame and be the sharpest element;
keep the environment subordinate and softly defocused.
Review the set side by side
Inspect all five images together:
Is it still the same product rather than a similar model?
Do color, texture, pattern, hardware, and structure match?
Is the color temperature and contrast coherent across the set?
Does the scene support rather than overpower the product?
Does the macro image resolve the material clearly?
Is there any text, watermark, or invented product claim?
Rerun only the failed image. There is no need to regenerate the entire set.
Troubleshooting
The set looks like five different products
The fidelity segment is too generic. Add color, material, hardware, stitching, pattern, and logo position, and use gpt-image-2.
Every image has a different color grade
Reuse the atmosphere segment word for word and add one shared grading instruction.
The macro image is soft
Use --quality high for that image and specify the exact texture units to resolve, such as leather pores, yarn plies, gear teeth, or marker edges.
The CLI returns unauthorized
Run dlazy login again or reset the API key.
The CLI returns insufficient_balance
Review and recharge dLazy credits:
dLazy Credits
Summary
The value of fission-pattern is not copying one image five times. It turns “keep the product fixed, vary the shot” into a repeatable production process:
Clean the source → Lock fidelity → Map the atmosphere → List five shots → Dry run → Generate and review
Once configured, the same recipe can be applied across multiple SKUs while keeping composition, background, and lighting consistent with the brand.
Related links
Original fission-pattern skill · dLazy CLI source · dLazy website
Top comments (0)