Two everyday problems, each starting from one photo: four hairstyles to compare before a cut, and a 500 yuan desk makeover with a render and shopping list.
1. Tasks and Results
Each task started with a single photo.
- Task 1: take one selfie, produce four hairstyle renders in different styles, then combine them into one comparison image to use as a reference for the barber.
- Task 2: take one photo of a desk and produce a makeover plan within a 500 yuan budget: a render of the result, an item image with numbered labels, and a shopping list you can order from directly.
Outputs of the two tasks:
- Task 1: one selfie produced four hairstyle renders and one combined 2×2 comparison image at 2480×3556.
- Task 2: one desk photo produced one makeover render, one image with nine numbered labels, and one shopping list of nine items totaling 500 yuan.
2. Task Breakdown
Task 1: Four Hairstyles From One Selfie
Background
A single front-facing selfie in natural light, with the hair currently a longer, fluffy textured crop. The point is to see the result before the haircut, rather than discovering only afterwards that it does not suit you. The four styles need to differ enough to compare side by side.
Approach
Step 1: choose the styles. The four options have to be spread out to cover the full range from conservative to distinctive, otherwise the comparison means nothing.
| No. | Style | Positioning |
|---|---|---|
| 1 | Korean textured perm | The middle option: forehead exposed, full and layered, suitable for any everyday setting |
| 2 | Short textured crop | Least effort: sides buzzed short, no styling needed after washing and drying |
| 3 | Side-part slick back | Business-leaning: faded sides, forehead fully exposed |
| 4 | Foil perm with wool curls | Most distinctive: tight curls all over, hair reads as thicker |
Step 2: generate each style. Pass the selfie to the generation step as a reference image and render one style per image. Two constraints matter: keep the facial features, face shape and skin tone unchanged and replace only the hair; output real photographic texture, avoiding an illustration or rendered look.
Step 3: compose the comparison image. Pass the four images to the image composition step and lay them out in a grid. The layout is fixed at two columns, each cell 1200×1600 pixels, with a circular number and the style name.
Results
| Output | Notes |
|---|---|
| Four hairstyle renders | 1536×2048 each, real photographic texture |
| Comparison image | 2480×3556, 2×2 grid, with numbers and style names |
Task 2: A Desk Makeover From One Photo
Background
A photo of an office desk: a dark grey desktop with a wood-grain divider, and nothing on it but a laptop, a mouse and a few loose items, with cables exposed. The budget is 500 yuan, and the result has to feel both cozy and cool.
Approach
Step 1: break down the item list. Decide what to buy within the budget first, then let the image present that list; doing it in the reverse order produces images and lists that do not match. The nine items fall into three groups by function: those that build the structure (monitor riser, desk mat), those that create atmosphere (ambient lamp, plant, art print), and those that contain the clutter (cable box, storage box).
| No. | Item | Function | Reference price (yuan) |
|---|---|---|---|
| 1 | Wooden monitor riser | Raises the screen, storage space underneath | 75 |
| 2 | Warm-light ambient desk lamp | The core of the cozy feel | 89 |
| 3 | Small potted plant + terracotta pot | The only living thing on the desk | 45 |
| 4 | Large leather desk mat | Frames the keyboard and mouse area | 55 |
| 5 | Cable box + cable clips | Hides the exposed cables | 29 |
| 6 | Wooden storage box + pen holder | Puts loose items back in place | 39 |
| 7 | Mug | Everyday necessity | 59 |
| 8 | Small Bluetooth speaker | The cool part | 74 |
| 9 | Art print | Fills the empty divider panel | 35 |
| Total | |||
| 500 |
Step 2: generate the render. Using the original desk photo as reference, keep the spatial structure unchanged: the same dark grey desktop, the same wood-grain divider, the laptop in the same position and at the same angle, with the nine items arranged into the scene. Real photographic texture required.
Step 3: make the labelled image. Overlay numbers 1 to 9 on the render at each item's position, matching the shopping list one to one so you can order while looking at the picture.
Results
| Output | Notes |
|---|---|
| Makeover render | 2048×1536, 4:3 |
| Numbered image | 2048 wide, nine numbers matching the list |
| Shopping list | 9 items, 500 yuan in total |
3. Where Talos Helps
In both tasks, the value of Talos lies in how the steps are chained together.
- The whole run happens in one conversation. The hairstyle task goes from reading the image to choosing styles, generating each one, and composing the grid, without leaving the conversation for another application. The desk task runs the same way: read the photo, break down the list, generate the render, overlay the numbers, and write out the shopping list file. Done separately, you would switch between the clipboard, an image editor and a spreadsheet tool several times.
- It handles the steps after generation too. Assembling the comparison grid and overlaying numbers on the render both reorganize images that already exist, which is image composition rather than generation. Only with both capabilities in the same place can you go from a single selfie straight to a comparison image you can take to the barber.
- The output is a file, not just a picture. The comparison image, the labelled image and the shopping list are all finished artifacts you can use as they are. The list is also saved as a file, so next time you change the budget and edit a few lines.
- Reference images keep the subject intact. The hairstyle task requires the facial features to stay the same while only the hair changes; the desk task requires the desktop structure and the laptop position to stay put. The renders were generated under those constraints, so only the parts that should change did.
4. Summary
Each of these tasks used one photo: one selfie produced four hairstyles plus a comparison image, and one desk photo produced a 500 yuan makeover plan plus a shopping list.
What they share is the same workflow: give it a reference image, state clearly what you want and what the constraints are, let it generate, then organize the generated images into a usable artifact. That holds for a haircut, for a desk makeover, and equally for a new profile picture, trying on clothes, or a color scheme for a room.
To try it, start with the simplest case: take a photo, write down what you want changed and what must stay, and hand it to Talos. Once you have run it through, you will find that the real time goes into thinking through the constraints, and that this step keeps paying off the more you practice it.
