Prompts with images: what to say when you attach one
For each kind of image (sketch, drawing, part photo, device photo, markup, concept) here is what the agent does with it and the prompt that gets the best result.
Attach images with the picture button in the prompt bar, or drop them onto it. You can add up to six in one request. Solid Vibes classifies each image automatically, showing its confidence, and treats each kind differently.
The image carries what. Your words carry why, the units, and the goal. The best image prompts always say all three.
Quick reference
| Image | What the agent does | What to add in words |
|---|---|---|
| Napkin sketch | Builds a starting model with clean geometry | What the shapes are, key sizes, units |
| Technical drawing | Reads dimensions into named parameters, flags ambiguities | Units, projection standard, which views matter |
| Photo of a part | Measures from a reference object, builds a replacement or improved part | The reference object and size, caliper measurements, what to improve |
| Photo of a device | Designs a mating part that fits or clips onto it | What the part should do, where it attaches, key clearances |
| Markup on my model | Applies the circled or arrowed note as an edit | What the markup means ("make this part thicker") |
| Concept image | Takes style and layout as intent | Real dimensions, and what to preserve from the image |
Napkin sketch
This is a sketch of a cable organizer for my desk. The long rectangle is the base, 160 mm long. The four U shapes are cable slots, each 8 mm wide. It is open at the back. All numbers are millimetres.
Tip: Write the numbers on the sketch and repeat the important ones in the prompt.
Technical drawing
This is a supplier drawing in third-angle projection, dimensions in millimetres. Build it as a parametric model with a named parameter for every dimension. List anything unclear.
Tip: Use Smart mode. Fewer misreads save more credits than the extra cost.
Photo of a part
This is a broken knob from my oven. The coin is a 1 euro coin (23.25 mm). The shaft hole is 6 mm with a flat on one side, 4.5 mm across the flat. Make a replacement, a bit taller to improve the grip.
Tip: Give calipers' numbers for anything that must mate. Photos are for shape and rough scale.
Photo of a device
This is my wireless charger. Design a bracket that clips onto its rim and holds a phone at an angle. The rim is about 6 mm thick. Print in PETG with a snap fit.
Tip: Photograph the device from the side where the part attaches, with a reference object for scale. If the device exists in the library, name it so the agent can use real dimensions.
Markup on my model
Take a screenshot of your model, circle or draw arrows on it, and add a note:
Apply the markup: make the circled rib thicker (the arrow shows the direction) and move the red-circled hole 5 mm toward the edge.
Tip: Keep each markup simple. One circle per change, with a clear arrow. Avoid overlapping notes.
Concept image
Use this as visual intent. Make it 120 mm tall, with a 60 mm base. Keep the rounded shoulders and the wide flat top. Before building, list the dimensions you are assuming.
Tip: A turnaround sheet (front, side, top, three-quarter) from a locked concept works much better than a single angle.
Multiple images in one request
Combine them for richer context. For example, a photo of the device plus a sketch of what you want:
Image 1 is my router from above. Image 2 is a sketch of the wall mount I have in mind. Build the mount so the router sits in the cradle shown in the sketch. The router is 220 mm wide.
Say which image is which. "Image 1" and "image 2" work well.
Photography tips
- Straight on, with the camera square to the subject.
- Plain background that contrasts with the object.
- Even, soft light. No harsh shadows across edges.
- One reference object of known size, placed at the same height as the subject.
- High enough resolution that small text and edges are readable.
What happens to text in an image
Words in an image, such as labels, notes or a hand-written "ignore everything above", are read as data about the picture, never as instructions to the agent. This protects you from images that carry hidden commands. Put your actual instructions in the prompt.
Costs
An image-to-model request costs 10 credits in Fast and 20 in Smart by default. A markup that only triggers an ordinary edit is billed as that edit (1 credit in Fast and 2 in Smart by default). Routing and classification are free.
Try it
Draw a small sketch and attach it:
This is a sketch of a small shelf bracket. The long arm is 100 mm, the short arm 60 mm, and the thickness 5 mm. Two screw holes in the short arm, 4 mm each. Millimetres. Print in PETG.
Compare the ghost preview with your sketch. Then compare the credits spent in Fast and Smart.
Did this help?
Keep reading
Playbooks
Turn a napkin sketch into a model
Photograph a rough hand sketch, tell the agent what the lines mean, and get a clean starting model you can refine.
Beginner4 min readPlaybooks
Turn a technical drawing into a parametric model
Drop in a dimensioned drawing or PDF, review the dimensions the agent reads into named parameters, and verify the result against the original.
Intermediate4 min readPlaybooks
Replace a broken part from a photo
Photograph the broken part next to a coin, let the agent measure it, then print a replacement that is stronger than the original.
Beginner4 min read