| 1 |
Flux -- network that creates images on text requests. |
| 2 |
Flux.2 [pro] -- best balance of quality and velocity, better start with it. |
| 3 |
Flux. 2 [max] -- maximal quality, most expensive. |
| 4 |
num of images -- a number of images per 1 generation. |
| 5 |
Prompt: 1.1. subject -- who or what have to be on photo. It helps AI what you need to image firstly. |
| 6 |
Prompt: 1.2. Style, details -- documentary or artistic, halftone effect, contrast, maximal or minimal elements of design. It's a visual style of image making necessary impression. |
| 7 |
Prompt: 1.3. Atmosphere -- pointing color gamma and mood, lighting, textured finish. |
| 8 |
Prompt: 1.4. Technical aspect -- reminding high resolution makes algorithm pay attention to tiny elements and details. |
| 9 |
Prompt: 1.5. Structure -- a well-structured description with maintaining order. Prompt is built logically (from common to partial). |
| 10 |
Prompt: 2.1. Subject -- who or what have to be on photo (human, animal or just object). |
| 11 |
Prompt: 2.2. Setting -- environment, location. |
| 12 |
Prompt: 2.3. Lighting -- how falls light (soft, hard, the golden hour). |
| 13 |
Prompt: 2.4. Style -- artistic or cinematic reference (vintage, documentary, the style of a certain director). |
| 14 |
Prompt: 2.5. Technical details -- camera type, focal length, depth of field, texture. |
| 15 |
Prompt: 2.6. Composition and mood -- how a frame is arranged, the atmosphere. |
| 16 |
Seed -- it's a number from that a model starts process of generation. |
| 17 |
Seed: using the same number and the same prompt a model every time will give you the same image. |
| 18 |
Seed: if left a field is empty, or typing a negative number or random then network woul give you every time different image, even using the same prompt. |
| 19 |
Seed: if you like result, you have to save seed (a number). It will be your base of image, but you can correct image via prompt (removing sunglasses and others). |
| 20 |
Steps -- a number of stages, that network doing during creation of image. |
| 21 |
The more Steps the more detailed and quality will get an image. But longer a process of generation. |
| 22 |
Steps can calle differently: iterations, sampling steps, inference steps. |
| 23 |
Steps for quality text at least you should put 40. |
| 24 |
Guidance -- it determines how better a model follows your prompt. If too high it can be anomalous proection because of your too technical prompt. |
| 25 |
Low number of guidance can give network more space of interpretation of your request. |
| 26 |
Guidance can call differently: guidance scale, prompt strength, CFG scale. |
| 27 |
Control_image -- for making image from reference. |
Комментарии