# Runway 參數指令推薦數值｜咒語寶典

來自 [@ShotbyEvan#2862](https://discord.com/channels/948250142350376990/1077654871416774676/1087456764636696667) 的整理

[回主頁](https://hackmd.io/@DCT/Runway-Gen1-Gen2-intro)

社群有人推薦的預設 prompt ：

> 影片差異度高 提高品質 文字權重重 流暢

--depth_blur_level 5 --cfg_scale 12 --interpolate --upscale

# 參數

原始提供者的文字內容

copy everything after '➠' until '--seed ##########' 
for customizable base remove 'TEXT PROMPT' 

WHAT PARAMETERS DO (this is Evans Understanding not 100%) --

Depth Blur = INPUT VIDEO INFLUENCE ON OUTPUT VIDEO '0-7' LOW NUMBER = LOOKS MORE LIKE THE INPUT VIDEO.

cfg temporal scale = AMOUNT OF GENERATION EACH FRAME / CONSISTENCY OF GENERATION ONTO VIDEO '.5 - 1.25' LOW NUMBER = DECREASED CONSISTENECY OR MORE GENERATIONS PER FRAME IN OUTPUT.

cfg scale = TEXT/IMAGE PROMPT INFLUENCE ON OUTPUT VIDEO '7.5 - 12.5' LOW NUMBER = LESS PROMPT INFLUENCE AND MORE RANDOM GENERATION.

Seed = THE GENERAL STYLE OF YOUR GENERATION - USE THE SAME SEED TO CREATE LONG FORM CONTENT.

Only Foreground Depth = ANIAMTES THE SUBJECT OF YOUR INPUT VIDEO AND REPLACES THE BACKGROUND WITH RANDOM GENERATED CONTENT.

mask modes = FOREGROUND ONLY ANIMATES SUBJECT LEAVING THE BACKGROUND OF THE INPUT VIDEO UNTOUCHED BACKGROUND DOES THE OPPOSITE.

Subsampling = THE NUMBER YOU PUT '1-5' IS THE AMMOUNT OF FRAMES SKIPPED 1 IS EXACT FRAME RATE OF YOUR INPUT VIDEO 5 = EVERY FIFTH FRAME OF THE INPUT VIDEO WILL BE GENERATED TURN ON INTERPOLATE IF SKIPPING LOTS OF FRAMES.

由 Bing Ai 翻譯成中文

複製 ‘➠’ 之後的所有內容，直到 ‘–seed ##########’。對於可定制的基礎，請刪除 ‘TEXT PROMPT’。

參數做什麼（這是個人非官方的理解，不是100％）-

深度模糊 = 輸入影片對輸出影片的影響 ‘0-7’ 低數字 = 更像輸入影片。

cfg時間比例 = 每幀生成量/影片生成一致性 ‘.5 - 1.25’ 低數字 = 減少一致性或每幀輸出更多生成。

cfg比例 = 文本/圖像提示對輸出影片的影響 ‘7.5 - 12.5’ 低數字 = 更少提示影響和更多隨機生成。

種子 = 您生成的一般風格 - 使用相同的種子創建長篇內容。

僅前景深度 = 將您輸入影片的主題動畫化並用隨機生成的內容替換背景。

遮罩模式 = 前景僅使主題動畫化，使輸入影片的背景保持不變；背景則相反。

子採樣 = 您放置的數字 ‘1-5’ 是跳過幀數量。1 是您輸入影片確切幀率；5=每第五幀輸入影片將被生成。如果跳過很多幀，則打開插值。

一些基本參數及特殊效果 僅供參考

發光/下雪/黏土動畫/綠幕動畫

⚫ Copy Paste Base -- 

  ➠ @Gen-1  --upscale

⚫ Customizable Base --

  ➠ @Gen-1  TEXT PROMPT 
--depth_blur_level 
--cfg_temporal_scale 
--cfg_scale 
--upscale False
--compare False
--interpolate False
--only_foreground_depth False
--green_screen False
--mask_mode_foreground False
--mask_mode_background False
--subsampling 
--denoising_steps 25
--seed 

⚫ Green Screen Base --
 
  ➠ @Gen-1  TEXT PROMPT --depth_blur_level 
--cfg_temporal_scale 
--cfg_scale 
--upscale False
--compare False
--interpolate False
--only_foreground_depth False
--green_screen True
--mask_mode_foreground False
--mask_mode_background False
--subsampling 
--denoising_steps 25
--seed 

⚫ Green Screen Base (Subject Animation) -- 

  ➠ @Gen-1  TEXT PROMPT --depth_blur_level 
--cfg_temporal_scale 
--cfg_scale 
--upscale False
--compare False
--interpolate False
--only_foreground_depth False
--green_screen True
--mask_mode_foreground True
--mask_mode_background False
--subsampling 
--denoising_steps 25
--seed 


⚫ Snow Effect --

  ➠ @Gen-1  Ice Sculpture in an Electronic Blizzard --depth_blur_level 2
--cfg_temporal_scale 1.0
--cfg_scale 8.5
--upscale True
--compare False
--interpolate False
--only_foreground_depth False
--green_screen False
--mask_mode_foreground False
--mask_mode_background False
--subsampling 
--denoising_steps 25
--seed 870496246 

⚫ Claymation (improved) --

 ➠ @Gen-1  Claymation insects --depth_blur_level 2
--cfg_temporal_scale 1.0
--cfg_scale 8.5
--upscale True
--compare False
--interpolate False
--only_foreground_depth False
--green_screen False
--mask_mode_foreground False
--mask_mode_background False
--subsampling 2
--denoising_steps 25
--seed 3500315098


⚫ Realistic Coral Reef --

 ➠ @Gen-1  coral reef --depth_blur_level 2
--cfg_temporal_scale 1.0
--cfg_scale 8.5
--upscale True 
--compare False
--interpolate False
--only_foreground_depth False
--green_screen False
--mask_mode_foreground False
--mask_mode_background False
--subsampling 2
--denoising_steps 25
--seed 3090379386


⚫ claymation --

  ➠ @Gen-1  Claymation Spider --depth_blur_level 2
--cfg_temporal_scale 1.0
--cfg_scale 8.5
--upscale True
--compare False
--interpolate False
--only_foreground_depth False
--green_screen False
--mask_mode_foreground False
--mask_mode_background False
--subsampling 2
--denoising_steps 25
--seed 3796971814

⚫ 3D Animated --

 ➠ @Gen-1  Kitten in front of a cyberpunk coral reef --depth_blur_level 2
--cfg_temporal_scale 1.0
--cfg_scale 8.5
--upscale True
--compare False
--interpolate False
--only_foreground_depth False
--green_screen False
--mask_mode_foreground False
--mask_mode_background False
--subsampling 2
--denoising_steps 25
--seed 868993799

⚫ 3d 2 --

➠ @Gen-1  3D render --depth_blur_level 2
--cfg_temporal_scale 1.0
--cfg_scale 8.5
--upscale True
--compare False
--interpolate False
--only_foreground_depth False
--green_screen False
--mask_mode_foreground False
--mask_mode_background False
--subsampling 2
--denoising_steps 25
--seed 3941104716

 ⚫ Claymation Sunlight --

  ➠ @Gen-1  Claymation Sunlight --depth_blur_level 2
--cfg_temporal_scale 1.0
--cfg_scale 8.5
--upscale True
--compare False
--interpolate False
--only_foreground_depth False
--green_screen False
--mask_mode_foreground False
--mask_mode_background False
--subsampling 2
--denoising_steps 25
--seed 588835966

⚫ Subject Claymation --

  ➠  @Gen-1  Claymation Spider --depth_blur_level 2
--cfg_temporal_scale 1.0
--cfg_scale 8.5
--upscale True
--compare False
--interpolate False
--only_foreground_depth False
--green_screen False
--mask_mode_foreground True
--mask_mode_background False
--subsampling 2
--denoising_steps 25
--seed 3796971814


⚫ Glow Effect 2 -- 

  ➠  @Gen-1  Crystal Skull --depth_blur_level 2
--cfg_temporal_scale 1.0
--cfg_scale 8.5
--upscale True
--compare False
--interpolate False
--only_foreground_depth False
--green_screen False
--mask_mode_foreground False
--mask_mode_background False
--subsampling 2
--denoising_steps 25
--seed 3835038273

⚫ 3D Fur Texture (underwater) --

   ➠ @Gen-1  Kitten in front of a cyberpunk coral reef --depth_blur_level 2
--cfg_temporal_scale 1.0
--cfg_scale 8.5
--upscale True
--compare False
--interpolate False
--only_foreground_depth False
--green_screen False
--mask_mode_foreground False
--mask_mode_background False
--subsampling 2
--denoising_steps 25
--seed 3514165655

 ⚫ Glow Effect --  

  ➠ @Gen-1  underground cave skeleton lit up by a green orb --depth_blur_level 2
--cfg_temporal_scale 1.0
--cfg_scale 8.5
--upscale True
--compare False
--interpolate False
--only_foreground_depth False
--green_screen False
--mask_mode_foreground False
--mask_mode_background False
--subsampling 2
--denoising_steps 25
--seed 803170213


⚫ Custom 1 --

➠ @Gen-1  shiny crystal spider --depth_blur_level 0
--cfg_temporal_scale 1.25
--cfg_scale 2
--upscale True
--compare False
--interpolate False
--only_foreground_depth False
--green_screen False
--mask_mode_foreground False
--mask_mode_background False
--subsampling 5
--denoising_steps 50



⚫ Custom 2 --

➠ @Gen-1  planetary biodome on the moon --depth_blur_level 3
--cfg_temporal_scale 1.25
--cfg_scale 8
--upscale True
--compare False
--interpolate False
--only_foreground_depth False
--green_screen False
--mask_mode_foreground False
--mask_mode_background False
--subsampling 5
--denoising_steps 50
--seed 3941104716


⚫ Custom 3 -- 

➠ @Gen-1  magnifying glass refracting light --depth_blur_level 5
--cfg_temporal_scale 1.25
--cfg_scale 8
--upscale True
--compare False
--interpolate False
--only_foreground_depth False
--green_screen False
--mask_mode_foreground False
--mask_mode_background False
--subsampling 5
--denoising_steps 50
--seed 3941104716

AI 生成 Gen2 參數

已失效，無法在 discord 使用

Welcome to the Runway Randomized Prompt Generator! Your task is to act as a prompt generator for text-to-video conversion. To begin, you need to understand the rules, parameters, structure of the prompt parameters, text reference generator, and walkthrough. Here are the details: RULES: Do not use codeblocks. Use the parameters as described. Always follow the response walkthrough. PARAMETERS: depth\_blur\_level: Determines the degree of similarity between the output and input video's structure. A higher value leads to more significant changes. only allow whole numbers for depth\_blur\_level. cfg\_temporal\_scale: Controls temporal consistency. Values below 1 decrease consistency, while values above 1 increase it. Recommended values are between 1.0 and 1.25. cfg\_scale: Enhances the influence of the text or image prompt. A lower value leads to more creative interpretation, while a higher value is stricter. seed: A randomly generated number between 0-999999999. upscale: Increases the output resolution with slightly longer runtime. compare: Shows the input and output videos side-by-side for comparison. interpolate: Adds frame interpolation for smoother playback. only\_foreground\_depth: Affects the foreground subject and generates content for the background. green\_screen: Replaces the background with chromakey green. mask\_mode\_foreground: Affects only the foreground subject and leaves the background untouched. mask\_mode\_background: Affects only the background and leaves the foreground subject untouched. subsampling: Controls drop frames. A value of 1 generates frames sequentially, 2 skips every other frame, and 5 is every 5th frame. Increase this value to 3 or 4 if your generation is ending too soon. denoising\_steps: The number of steps used to denoise the final output. Values above 50 have minimal effect. STRUCTURE OF THE PROMPT PARAMETERS: Start with "@Gen-1" followed by randomized parameters from the list above. It is not necessary to use all of them. The user's response should impact the influence of the generated prompt. Example of parameter section of prompt: @Gen-1 --cfg\_temporal\_scale 1.2 --cfg\_scale 10 --seed 123456789 --interpolate TEXT REFERENCE GENERATOR: Provide suggestions for the art style for the video in text-to-art format. Include a specific artist’s style. Use colorful, artistic language to describe the video’s theme, including any specific visuals or moods you want to convey that best relate to the video described. Any text reference suggestion should be between brackets and be less than 35 words. Example of text reference for a video of a mountain: \[Impressionistic landscape painting in the style of Claude Monet, with bold brushstrokes and a soft color palette. The focus is on the grandeur of the mountain and the play of light and shadow on its surface\] WALKTHROUGH: Print "Runway Randomized Prompt Generator". Skip a line. Ask "What kind of video will you be using for Runway’s Gen-1? Will you be using an image reference or a text reference? If using a text reference, do you have any suggestions for the artistic theme being generated?" Wait for the user's response. If the user responds that they will use an image reference, provide a randomized list of parameters as described in the “Structure of the prompt parameters” section. If the user responds that they will use a text reference, provide a randomized list of parameters, skip a line, and create three randomized examples as described in the “Text reference generator” that would suit the video described.每次回覆最後都加上該回覆內容中每個 prompt 的中文翻譯 Now that you have all the information you need, let's begin