Furry monster high-tops, bunny-ear platforms and one-eyed dad shoes: how Talos turned one prompt into 8 shoe concept images and 3:4 posters in 3 minutes.
Project background
Producing a set of "shoe concepts + vertical covers" by hand usually takes 2-3 days: gathering references, settling on a style, drawing or rendering concept images, then re-laying out a poster for each aspect ratio. With Talos, this run produced 8 directly usable images in 3 minutes.
*01 Submitting the request and getting 4 directions*
Challenge
Shoe visuals are hard for two reasons. First, the concept stage has to show several directions at once, and hand-drawing or modeling each direction is a full round of work — a day is often only enough to polish one design. Second, the same shoe has to be produced in two compositions: a 1:1 product spec sheet on white and a 3:4 vertical cover, which differ in aspect ratio, negative space, and background. Done manually, that means re-laying out the design, and changing the colorway or style often comes close to starting over.
The goal here was to run shoe concepts and social covers through one workflow, so the output could go straight into the next stage of asset production and marketing.
*02 Using the final design as a reference to generate 4 posters*
Method
Step 1: Put style, elements, and specifications into the prompt
The first round used four prompts in the same session, each for a distinct design direction. The text of one of them:
A product image of an originally designed ugly-cute style high-top sneaker. The upper is wrapped in off-white fluffy long plush; a pair of pointed plush ears stand up on both sides of the tongue; a tuft of erect fur sticks up on the forehead; a single half-open round eye and two small exposed fangs hang on the front; a small plush tail trails from the side; the laces are replaced by a colorful beaded chain. Thick, rounded cream-colored rubber outsole with a subtle paw-print emboss on the side. Overall a designer toy art toy textured 3D render, matte soft vinyl mixed with plush, chibi exaggerated proportions, cartoon fine outlines, gender-neutral style, wearable by both men and women. Pure white seamless background, soft studio lighting, product still-life photography, 45-degree front-side view, subject centered, square composition, clean image with no text.The other three used the same approach, swapping material, silhouette, elements, and colorway to yield four non-repeating directions. This prompt contains four key ingredients:
Audience and style anchor: state that it targets young people, in an ugly-cute style, gender-neutral, so the model does not have to guess the aesthetic direction. Drawable details: plush, soft vinyl, big round eye, small fangs, thick sole, paw-print emboss — every element can land directly on the image. Color constraints: off-white with cream, fluorescent green with cream white, violet with black plush, so the four directions do not collide on color. Output specification: 1:1 square, pure white background, product still-life photography, so the results already meet the size requirements for proposal presentation and e-commerce.
The second round moved to posters, and the approach changed: one selected design was used as a reference image, and the prompt shrank to a single sentence.
Use this pair of furry bunny-ear ugly-cute platform shoes to generate a vertical promotional poster that attracts young people to buyStep 2: Talos generates automatically
- Parses the four prompts into four differentiated design directions, each with its own material, silhouette, elements, and colorway
- Generates four 1:1 white-background shoe design images at 2048×2048
- The designer picks the usable directions; this run selected the long-eared bunny platform shoe and the one-eyed horned dad shoe
- Uses the selected design as a reference image, locks the shoe shape and colorway, and switches the composition to 3:4 vertical
- Generates four vertical promotional posters at 1536×2048
- All eight images are stored in the session directory, ready to download at any time
The whole run took about 3 minutes.
One intermediate round is worth recording: the first poster version wrote composition, negative space, light position, and copy placement all into the prompt, and the results were not good. Cut down to one sentence, keeping only "which shoe, what ratio, for whom", all four came out usable.
Step 3: Output structure
The four shoe design concepts vary across four variables: material, silhouette, elements, and colorway:
| Concept | Visual focus | Main colors |
|---|---|---|
| Furry monster high-top boot | Long plush over the whole shoe, pointed ears, bared teeth | Off-white with cream |
| Long-eared bunny platform sneaker | Bunny ears grown into the tongue, blue-purple gradient | Cream white with light purple |
| Grinning big-mouth clog | One-piece molded big mouth, detachable eyeball charm | Fluorescent green with cream white |
| One-eyed horned dad shoe | Single eye and pointed horn, soft vinyl mixed with plush, jagged platform | Violet with black plush |
The four vertical posters reuse the two selected shoes and cover two kinds of placement:
| Poster | Design direction | Best placement |
|---|---|---|
| Bunny-ear shoe promo poster | Standard product promotion, emphasizing the bunny ears and platform | Note cover image |
| One-eyed dad shoe promo poster | Standard product promotion, emphasizing the big eye | Note cover image |
| Bunny-ear shoe street poster | Street style, downplaying the product-display feel | Content image |
| One-eyed dad shoe minimalist poster | Minimal composition, enlarged single eye | Series cover |
*03 Finished posters at 3:4, ready to use as covers*
Result
| Metric | Manual production | Talos | Change |
|---|---|---|---|
| Total time | 2-3 days (estimate) | About 3 minutes | About -99.9% |
| Directions in the first round | One image per direction, one at a time | Four differentiated directions at once | — |
| Changing ratio for posters | Re-layout and rebuild the background | Regenerate from a reference image plus a one-sentence prompt | — |
| Output format | Depends on the tool; often export and then crop | PNG images, 2048×2048 and 1536×2048 | — |
| Reusability | Start from zero each time | Prompts and reference images saved and reused | High |
Talos produced 8 images in about 3 minutes: four 1:1 white-background shoe design concepts and four 3:4 vertical posters. The poster ratio matches the Xiaohongshu cover, so no cropping is needed after upload. The time figure is based on feedback from the person who ran it.
Business impact
Output per run: a set of shoe concepts plus matching covers, compressed from a two-to-three-day schedule into a single session.
Repeat generation: once style and specifications live in the prompt, a new shoe only requires swapping the element descriptions, so one product line can expand quickly to multiple SKUs.
Style consistency: using the final image as a reference for further generation keeps the shoe shape and colorway from drifting between the concept sheet and the poster.
Delivery format: the output is standard-size image files that can go straight into Xiaohongshu, a product detail page, or a conversation with a sampling partner.
Replicable takeaways
- Keep prompts simple: for a single-image task, a reference image plus one sentence is enough. Writing composition, light position, and negative space all out in full detail tends to make the output worse.
- Specify up front: state the ratio, background, and whether to leave negative space in the prompt, and skip the cropping and cutout work later.
- Explore wide, then narrow: ask for four differentiated directions in the first round — it costs very little; once one is chosen, build the matching assets around it.
- Keep using the final image as a reference: from design sheet to poster, from product image to cover, the same reference image locks the style in place.
FAQ
Q: Can the generated images be used directly as Xiaohongshu covers?
A: Yes. The posters from this run are 1536×2048 at 3:4, the same ratio as a Xiaohongshu cover, so no cropping or border is needed after upload.
Q: Do the prompts have to be very long?
A: No. The shoe design round needs material, elements, colorway, and specifications spelled out; at the poster stage, put the final image into the reference slot and keep the prompt to one sentence — the results are steadier than with a long prompt.
Q: Can someone who cannot do 3D modeling or draw use this?
A: The input for this workflow is a written description plus one final image; no modeling or drawing software is involved, and the output can go directly into design discussions and online assets. Finer production details still need to be added by a designer or a sampling partner.
Q: Can this method be reused for other design subjects?
A: Yes. Just swap the subject and material descriptions; apparel, bags, e-commerce product images, and event posters all apply, and the reference-image step is especially effective for apparel and designer toys.