{"id":27233,"date":"2025-01-17T11:26:17","date_gmt":"2025-01-17T03:26:17","guid":{"rendered":"https:\/\/www.1ai.net\/?p=27233"},"modified":"2025-01-17T11:26:17","modified_gmt":"2025-01-17T03:26:17","slug":"%e5%a6%82%e4%bd%95%e7%94%a8%e5%9b%be%e7%89%87%e5%8f%8d%e6%8e%a8%e6%8f%90%e7%a4%ba%e8%af%8d%ef%bc%9f%e6%9d%a5%e8%af%95%e8%af%95%e7%94%a8ai%e5%8f%8d%e6%8e%a8%e7%94%a8%e5%9b%be%e7%89%87%e7%94%9f%e6%88%90","status":"publish","type":"post","link":"https:\/\/www.1ai.net\/en\/27233.html","title":{"rendered":"How to backpropagate cue words with images? Let's try generating descriptors with images using AI backpropagation"},"content":{"rendered":"

You see a super cool picture online and think, \"Wow, that's awesome, how can I get a similar one?\"<\/p>\n

At this time, the Backpropagation Image Cue Words technology is like your private detective, helping you analyze this image and find out the secrets behind it - that is, those keywords that may be used to generate this image. Then, you can tweak the original image to create your own style, or let AI help you draw a picture that is closer to your idea~.<\/p>\n

Compare 5 of the hottest cue word backpropagation tools to see the results together.<\/p>\n

First of all, the demonstration chart that we uniformly use is the following, which I generated in Instant Dreams, and the cue words after backpropagation are also uniformly generated in Instant Dreams.<\/p>\n

\"b6ef8a72j00sq7qm90027d000g000sgm\"<\/p>\n

I. Tencent Yuanbao<\/p>\n

Website:https:\/\/www.1ai.net\/11784.html<\/a><\/p>\n

First, ask a question after uploading the demonstration image thatYou are an expert in AI painting cue words, may I ask, what style of image is this?<\/strong><\/p>\n

\"2747b454j00sq7qmc008cd000u000fwm\"<\/p>\n

Then, ask the following questions in turn (no need to upload images anymore):<\/p>\n

Please describe all the details of this painting without missing a single detail.<\/p>\n

Please combine all the details into one paragraph and turn it into an AI painting cue word for me.<\/p>\n

Very good! Please write me 5 more similar sets of AI painting prompt words.<\/p>\n

\"508c5371j00sq7qmb002vd000u000k8m\"<\/p>\n

Then you get the following cue words, and after you get these, go to i.e. Dream and generate them, with the following results.<\/p>\n

\"bb417696j00sq7qmb00mud000u000fdm\"<\/p>\n

Found that the effect is far from the original picture, mainly because the prompt word is too general, so this is not recommended to use the Tencent Yuanbao counterpropagation.<\/p>\n

II. A Thousand Questions on the General Meaning<\/p>\n

Website:https:\/\/www.1ai.net\/6763.html<\/a><\/p>\n

After uploading the image, ask questions.Analyze this image carefully, I'm going to draw the same image using AI drawing tools, help me write the prompts<\/strong>.<\/p>\n

\"c6e218d1j00sq7qmb003bd000u000fqm\"<\/p>\n

Then using these cue words, the following results were generated.<\/p>\n

\"9b334c5cj00sq7qmb00i8d000u0009fm\"<\/p>\n

The generated effects, except for the first one, are not passable, so I don't really recommend the Tongyi Thousand Questions.<\/p>\n

III. Wisdom and Clear Words<\/p>\n

Website:https:\/\/www.1ai.net\/2443.html<\/a><\/p>\n

After uploading the image, ask the following question.Please analyze all the elements you see in this image, including color, composition, content, style, etc. and output it to me in a paragraph, being careful to include all the elements and not leave them out.<\/strong><\/p>\n

Then, continue with slight modifications and questions to get the following keywords.<\/p>\n

\"e9a6993ej00sq7qma003bd000u000bfm\"<\/p>\n

The generated result is as follows, and the robot is not metallic. So, the generated one is not so good either.<\/p>\n

\"0673ad0ej00sq7qmc00svd000u000dwm\"<\/p>\n

IV. chatGPT<\/p>\n

Website:https:\/\/www.1ai.net\/10491.html<\/a><\/p>\n

Upload an image and add the cue word thatPlease talk about all the elements you see in this image and analyze them, including color, composition, content, style, etc., and output them to me in a paragraph, taking care to include all the elements and not leave them out.<\/strong><\/p>\n

\"d838d0c6j00sq7qmb008hd000u000fsm\"<\/p>\n

Then, generate the cue word thatCould you please take this passage and summarize it into cue words that can be used to draw with Ai.<\/strong><\/p>\n

\"6fdd1252j00sq7qmb0050d000u000ism\"<\/p>\n

OK! After we get the cue word, go to that after the dream generation, the effect can be regarded as the upper middle, not bad.<\/p>\n

\"eb60d777j00sq7qmc00uyd000u000f7m\"<\/p>\n

V. Midjourney<\/p>\n

Website:https:\/\/www.1ai.net\/10463.html<\/a><\/p>\n

This is one I tried this afternoon specifically with a Midjourney membership. The way to use it is, use the demo image + \/describe command.<\/p>\n

\"b354cfa7j00sq7qmc00cmd000u000r7m\"<\/p>\n

Then, 4 commands come out directly. Click on 1, 2, 3, 4 below to generate images corresponding to the serial numbers of the commands.<\/p>\n

\"67f7bd0fj00sq7qmc00aqd000u000mvm\"<\/p>\n

Then I took 1,2,3,4 and tested them in my dream, and the result of 3 is the closest to the style of the demonstration diagram.<\/p>\n

\"f70a0a32j00sq7qmc00sld000u000eim\"<\/p>\n

\"66830f2bj00sq7qmc00tud000u000fbm\"<\/p>\n

\"c0e32741j00sq7qmb00jod000u000asm\"<\/p>\n

\"03a5ade3j00sq7qmb00jid000u000afm\"<\/p>\n

Then I tested the lower generation effect in midj itself! It's pretty good too.<\/p>\n

\"47a415a6j00sq7qmc00g1d000u000o3m\"<\/p>\n

Looking at the generation of this chart in instant dream, I personally think chatGPT and Midjourney have the best results. Domestic performs average.<\/p>\n

However, there is inherently some randomness in the outturn, so if you find that it doesn't quite fit, you can do so by manually modifying the cue words after backpropagation, or by padding the chart.<\/p>\n

What do you think?<\/p>","protected":false},"excerpt":{"rendered":"

You see a super cool image online and think to yourself, \"Wow, that's an awesome image, how can I get a similar one?\" At this point, the Backpropagation Image Cue Words technology acts like your private investigator, helping you analyze the image and find out the secrets behind it - that is, those keywords that might have been used to generate the image. This way, you can make adjustments to the original image to create your own style of image, or let AI help you draw a picture closer to your idea ~ compared the 5 hottest cue word backpropagation tools, to see the effect together. First of all, we unify the use of the demonstration map is as follows, I generated in the dream, after the reverse thrust of the prompt word is also unified in the dream generation. First, Tencent Yuanbao Website: https:\/\/www.1ai.net\/11784.ht<\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[144],"tags":[5575,4058,914],"collection":[],"class_list":["post-27233","post","type-post","status-publish","format-standard","hentry","category-baike","tag-ai"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/27233","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/comments?post=27233"}],"version-history":[{"count":0,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/27233\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/media?parent=27233"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/categories?post=27233"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/tags?post=27233"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/collection?post=27233"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}