1
Starting input
Prompt-led creation
A written description of the subject, setting, style and composition.
Reference-led editing
An existing image plus instructions describing what should change or remain.
Image generation
Explore novita image models for your next visual
An image model turns a written prompt, a reference image, or both into a visual result. This guide separates prompt-led creation from reference-led editing so you can choose an input that fits the result you need. Model-specific inputs and controls should be checked before you begin.
Why A-to-B
The most useful distinction is what you supply first: a description of a new scene or an existing image to preserve and change. Availability of either workflow depends on the selected model.
Prompt-led creation
Reference-led editing
2
Best fit
Prompt-led creation
Exploring a scene when you do not need to match an existing picture.
Reference-led editing
Revising a supplied visual while retaining recognizable elements.
3
Continuity
Prompt-led creation
Describe recurring details explicitly; each new result may reinterpret them.
Reference-led editing
Use the reference to communicate details that words alone might leave ambiguous.
4
Prompt emphasis
Prompt-led creation
Specify the subject, camera viewpoint, lighting and desired visual treatment.
Reference-led editing
Identify the edit and state which parts of the source must stay intact.
5
Review focus
Prompt-led creation
Check whether the scene, objects and composition match the brief.
Reference-led editing
Check both the requested change and unintended changes to the source.
6
Before selecting a model
Prompt-led creation
Confirm that it accepts a text prompt and returns the output you need.
Reference-led editing
Confirm that it accepts an image input and supports the kind of edit you intend.
The tool block
Start with the image you need, then choose a model that accepts your available input. These scenarios show where a still-image workflow may connect to a separate voice workflow.
Storyboard artist
Describe a scene, generate a still, and compare its framing against the shot list before making another pass.
If the storyboard later needs spoken dialogue, novita voice models covers a separate audio workflow.
novita voice modelsLesson designer
Create an illustration for a concept, then check that its visual details support the written explanation.
For an accompanying spoken lesson, novita voice models addresses voice output rather than image generation.
novita voice modelsProduct communicator
Test a visual direction with a precise prompt, or use a reference when an existing composition must be retained.
When the presentation also calls for narration, novita voice models is the related audio-model guide.
novita voice modelsSpec table
Choose an input, then inspect the output
Use the comparison above as a checklist, not a promise that every image model accepts every input. Write down what the visual must show, choose a compatible workflow, and review the result for missing objects, unwanted changes and details that matter to your project. Refine the instruction when the output misses the brief.
- Confirm supported inputs for the model you select
- Describe the visual goal in concrete terms
- Review the output before sharing it
Variant FAQ
Image models are used to create or modify visuals from supported inputs. A written prompt can describe a new scene, while some workflows may accept an existing image as a reference or editing input. Check the selected model’s input requirements rather than assuming that all image workflows work the same way.
Start with a prompt when you want to explore a visual without preserving a particular source. Start with a reference when recognizable features, composition or other source details are important. First confirm that the model you plan to use supports image input.
Name the subject and describe the setting, viewpoint, lighting and visual style that matter most. If a detail must appear, state it directly instead of relying on the model to infer it. Review the result and revise the prompt around the specific detail that missed the brief.
Do not assume so. Image generation and image editing can require different inputs or model capabilities, even when both produce an image. Look at the selected model’s supported input types before preparing an editing task.
Compare it with the original brief and inspect important objects, text, spatial relationships and any details carried over from a reference. A visually appealing image can still miss a required feature or alter something you intended to keep. Make another pass when those checks fail.