Jump to content

Media creation with artificial intelligence: Difference between revisions

m
no edit summary
mNo edit summary
mNo edit summary
Line 73: Line 73:
'''Beginner workflow'''
'''Beginner workflow'''


Mini tutorial based on ChatGPT. Though, this should similar for all MLLMs.
Mini tutorial based on ChatGPT:
* '''Iterative prompting''': Describe the desired result as clearly as possible: Motive, perspective, colors, lights, shadows, art style. Refine the prompt step by step based on undesired aspects rather than expecting a perfect result on the first attempt. Negative prompts: You can also explicitly write what you don't want.  
* '''Iterative prompting''': Describe the desired result as clearly as possible: Motive, perspective, colors, lights, shadows, art style. Refine the prompt step by step based on undesired aspects rather than expecting a perfect result on the first attempt. Negative prompts: You can also explicitly write what you don't want.  
* '''Avoid quality loss''': If the GenAI degenerates the image quality because of too many iterations, try from a new start with combined text prompts.
* '''Avoid quality loss''': If the GenAI degenerates the image quality because of too many iterations, try from a new start with combined text prompts.
Line 80: Line 80:
* '''Merging / composition''': Supplying multiple images in a single prompt can help combine elements. However, repeated re-editing of generated outputs may degrade detail or introduce artifacts.
* '''Merging / composition''': Supplying multiple images in a single prompt can help combine elements. However, repeated re-editing of generated outputs may degrade detail or introduce artifacts.
* '''Batch generation''': Since outputs are probabilistic, generate multiple final candidates and select the best.
* '''Batch generation''': Since outputs are probabilistic, generate multiple final candidates and select the best.
* '''Post-processing workflow''': Combine the best elements using external tools (e.g., masking in Photoshop or Krita). This hybrid approach often yields higher-quality results than relying on a single generation.<!--Not for these MLLMs:
* '''Post-processing workflow''': Combine the best elements using external tools (e.g., masking in Photoshop or Krita). This hybrid approach often yields higher-quality results than relying on a single generation.<!--Not for chatbots:
* '''Consistency strategies''': When available, use features such as seeds, style references, or controlled variations to maintain visual coherence across multiple images.-->
* '''Consistency strategies''': When available, use features such as seeds, style references, or controlled variations to maintain visual coherence across multiple images.-->


8,844

edits