Reference images: when to use them, what size, how the app sends them

A reference image is the quickest way to make Dola keep one character, one product or one setting across many videos. It is also the most common reason for a refusal. This article helps you choose the right images and explains what Spider Hub does with them before sending.

When to use images

Use them when:

• You need the same character in several shots.
• You're making videos for one specific product (a bottle, a box, a pair of shoes) and want Dola to keep its shape and colours.
• You have a setting or look you've settled on (a room, a colour palette) that has to repeat.

Skip them when:

• The video only needs a general mood ("a rainy street at night", "a beach at sunset"). Words are enough, and that's one less reason for Dola to refuse.
• The only image you have is a photo of a real person. See below.

Adding images to a shot

• On a draft card, click the + button to add images. Each shot takes up to 30 images, as PNG, JPG, JPEG or WEBP.
• To add the same set to every draft card (say, a whole episode with the same characters): Studio's Tools menu → Add reference images to every draft…
• The app skips an image whose content is already in the same shot.

Order matters. Dola numbers images in upload order: image 1, image 2… Your prompt should name each image in the first sentence, as in A prompt formula for 30-second videos:

图1中的角色是一位名叫小陈的年轻女背包客，图2中的角色是一只名叫阿福的橘色小狗。

Choosing good images

• One clear subject per image. A character image shows one character, a product image one product, on the plainest background you can manage.
• Bright and sharp. A blurry or dark image gives Dola little to hold on to.
• No need for huge files. About 2048 px on the long side carries enough detail. 4K images still work; they just upload more slowly.
• The same set of images for the same character, in the same order, in every shot.

Real people and portrait protection

Dola protects portraits: with a real person's face in an image, Dola usually only makes videos of the account owner themselves. Other real people are commonly refused, and the shot's card is titled Reference image shows a real person.

Dola doesn't say which image it means. If the shot has one image, the card names it as @IMAGE_1. With several, the card says plainly that Dola didn't specify, and suggests trying again without the images that show a real face.

What to do:

1. Click Edit and run on the card.
2. Remove images with real faces and use an illustrated character, a product, a place, or words only.
3. Generate again.

How the app sends heavy images: original first, light copy after

An image longer than 2048 px on its long side, or a file over 2.5 MB, counts as heavy. For heavy images Spider Hub goes in this order:

1. It sends the original file first, so Dola gets the best quality. The app waits patiently for the upload, up to about 3 minutes per try, and tries on a few fresh creation pages if it gets stuck.
2. If the original still sticks, times out or Dola rejects the file (for being too large, say), the app sends a light copy: at most 2048 px on the long side, correctly rotated, with transparency kept for transparent images.
3. If the light copy won't upload either, the shot stops before the prompt is sent, so nothing is charged. The card says Reference image error and suggests Edit and run to replace the image.

Once the original has failed for a shot, later new chats for that shot send the light copy straight away rather than making you wait all over again.

Your original file is never changed. The light copy is a separate copy in the app's temporary folder.

Images that aren't heavy (under 2048 px and under 2.5 MB) are sent as they are.

Example: videos for one product

You're making 10 fifteen-second videos for a perfume bottle:

1. Photograph the bottle on a plain background in good light: one front view, one side view, one close-up of the cap.
2. Paste 10 prompts with Paste Batch, each in a different setting (dressing table, beach, hotel room…).
3. Tools menu → Add reference images to every draft… → choose the 3 images.
4. Start every prompt with a sentence saying images 1, 2 and 3 are the same perfume bottle, then describe the action and the light.
5. Generate all.

No real people in the images, the product keeps its shape, and all 10 videos share one set of images.

Summary

• Use images to keep a character, product or setting across shots; skip them for mood alone.
• Up to 30 images per shot; PNG, JPG, JPEG, WEBP; image order is Dola's numbering.
• Images with real faces are often refused; the app never works around that filter.
• Heavy images: the original goes first, then a light copy of at most 2048 px; your original is never changed.
• If an image can't upload at all, the shot stops before the prompt is sent and nothing is charged.