How to write prompts for Sora, Midjourney and Kling: a guide for beginners

Briefly about the main thing (BLUF)

A good prompt has a formula: subject, action, style, camera, light - and you shouldn’t deviate from it. The guide breaks it down by task: pictures (Midjourney, Flux), video (Sora, VEO), image-to-video (Kling) - with typical problems that I went through myself.

Are you familiar with the situation: you see someone’s amazing result in Midjourney, you copy a similar prompt - but the result is nonsense? I've been through this. It turned out that the problem is not in the model, but in HOW you formulate the request. 80% of the result is a prompt. Here are the rules that actually work for Sora, Midjourney, Kling and in general any generative model. We use them internally NeuralSpace.

Basic formula - remember and don't give up

Subject + action + environment + style + light + composition.

Each link improves controllability. Without a subject, the model will come up with something of its own. Without style - it will give an “average photo”. Without indicating the light, you will lose the atmosphere. It is not necessary to register all six every time, but the more, the more accurate the hit.

Prompts for pictures (Midjourney, Flux)

• Short blocks separated by commas: “girl in a red coat, autumn park, soft backlight, 35 mm film”
• Indicate style and references. “Watercolor”, “isometric”, “in the spirit of Wes Anderson” - the model understands this
• Negatives - separately! Do not write “no text” in the main prompt, there is a negative prompt for this
• IN NeuralSpace you can write in Russian - the platform itself will translate and optimize

Prompts for video (Sora, VEO)

• First the frame, then the movement: “a city street in the evening, the camera slowly moves backwards, people with umbrellas pass by”
• Specify the duration: “3 seconds”, “short movement”. Without this, the model will decide on its own and may not guess
• Difficult video? Break into scenes - Sora 2 Pro Storyboard in "Video" section This is what it was created for

Image-to-video (Kling)

The main mistake here is to describe the picture again. The picture is already there! Describe only the movement: “hair flutters in the wind from the left,” “the character takes a step, the camera is static.” Brief and specific.

Typical mistakes (I did everything)

• Prompt for 200 words - the model gets confused and produces a mess
• Contradictions: “winter night, sunshine” - seriously, people write that
• Five references in one request. One or two - maximum
• Trying to fix everything in one prompt instead of step-by-step refinement

Templates and successful prompts can be seen in "Trends" section — there you can see what other NeuralSpace users generate.

Frequently asked questions (FAQ)

Question: How to get the best result from a neural network?
Answer: Use detailed prompts (descriptions) in English, set the style and details of the scene.

Question: Can these materials be used for commercial purposes?
Answer: Yes, the generated content is entirely yours.