Correction, 2026-10-01: an earlier version described a six-element Veo formula as one "used internally" at 8frame and said Nano Banana is optimized and trained for product photography and returns a product photo "in seconds." There is no internal formula, we do not know what Nano Banana was trained on, and on 8frame its measured median generation time is 26 seconds (Nano Banana 2). The formulas below are our suggestions, not model rules. It also implied that only the video tools lack a negative-prompt field; no image or video tool on 8frame has one.
Prompt engineering is the practice of writing and structuring text inputs to reliably produce high-quality outputs from a generative AI model.
The term sounds more specialized than it is. At its core, it's just figuring out how to talk to a model so it does what you actually want. Every AI model, whether it generates images, video, or text, interprets your input and produces output based on patterns it learned during training. A vague input produces a vague output. A precisely structured input gives the model the context it needs to make good decisions about subject, style, motion, lighting, and framing. Prompt engineering is the skill of providing that context deliberately.
How prompt engineering works
Models don't reason the way humans do. They produce the output their training makes most likely given your input. That means how you phrase something matters as much as what you're asking for.
A few habits that help across most generation models:
Specificity beats brevity. "A product photo" gives the model almost no signal. "A glass perfume bottle on a white marble surface, soft side lighting from the left, shallow depth of field, studio photography" gives it dozens of signals to constrain the output space.
Lead with the subject. Put the core subject first, then camera or lighting, then style or mood modifiers. It keeps prompts readable and makes it easier to see which part to change when a take misses.
Say what to avoid, where the tool allows it. Some models accept a separate negative prompt. On 8frame's image and video tools there is no negative-prompt field, so describe the look you want instead ("crisp, clean frame") rather than listing what you don't want.
Iterative refinement. Prompt engineering is rarely one-shot. You run a prompt, read the output, identify what the model misunderstood, and adjust. The gap between your first attempt and your tenth is usually visible.
When you use prompt engineering
Any time you use a generative AI tool, you're doing some version of this. The difference between casual use and intentional prompt engineering is whether you're conscious about structure.
You'll invest the most effort in prompts when:
- You're generating assets for a client or campaign with a specific visual brief.
- You're building a repeatable workflow where consistent output matters across dozens of runs.
- You're working with a new model you haven't used before and need to learn its tendencies.
- You need to match an existing visual identity: a brand's color grading, a character's face, a specific lighting setup.
Examples in image and video generation
Veo 3.1 on 8frame. A useful checklist for video has six elements: subject, environment, camera movement, lighting, film style, audio cue. An example for a product clip:
"Glass sneaker on a black volcanic rock surface, slow low-angle push-in, dramatic side lighting with hard shadows, 35mm film grain, ambient outdoor wind sound"
Each element constrains a different dimension of the output. Leave out the camera movement and the model chooses one for you; leave out the lighting and you get whatever it defaults to. The audio cue matters because Veo 3.1 generates sound with the clip by default. On 8frame the Veo tool starts on Veo 3.1 Fast, 8 seconds with sound, 168 credits.
Nano Banana on 8frame. For product stills, a three-element formula is a good start: Product + Surface + Lighting.
"Matte black water bottle, weathered concrete countertop, diffused window light from above"
Name the light the way a photographer would ("diffused window light from above") instead of describing it loosely. Nano Banana 2, the default, is 11 credits an image at 1K, returns up to four images a run, and has a median generation time of 26 seconds on 8frame.
Both formulas work the same way: they use the vocabulary of photography and film, which these models have seen described in their training data. Prompt engineering is partly vocabulary acquisition: learning which terms a given model responds to and how.
FAQ
Is there a negative prompt on 8frame? No. None of the image or video tools (Veo, Kling, Nano Banana and the others) has a negative-prompt field. Describe the result you want instead.
Do longer prompts cost more on 8frame? Not for image and video generation; price depends on the model and its settings. Text-to-speech is the exception, billed per started 1,000 characters.
Related concepts
- How to write Veo 3 prompts goes deeper on structuring Veo prompts, with examples across scene types.
- Nano Banana vs Seedream vs FLUX compares how the image models on 8frame respond to prompts, which shows how prompt engineering has to adapt per model.
Sources
- Veo and Nano Banana node inputs, defaults and prices; ElevenLabs TTS billing: 8frame tool registry, read 2026-10-01
- No negative-prompt field on any image or video node: 8frame front and back tool registries, checked 2026-10-01
- Nano Banana 2 median generation time (26 s, completed tasks, last 120 days): 8frame production data via CONTENT_FACTS, verified 2026-09-30
Want to put this into practice? Open the 8frame canvas, run the same prompt on two models side by side to compare how each one interprets it.
Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.