8 months ago, I compared the text generation abilities of AIGC tools Dall E2 and Mid Journey. I intentionally challenged them with ancient poetry. At that time, my judgment was that Dall E was more suitable, although the quality of the generated images was slightly lower, it had a better understanding of the prompt.
Furthermore, I also evaluated which AIGC engine to use in our product, and ultimately chose to use Dall-E.
Now Dall-E 3 is out (currently still need to obtain it through Microsoft's Bing, but it is expected to be integrated through API in October, so that ArcBlock's products can also be fully upgraded to support Dall-E 3), let's test it again.
Not surprised, directly translating the whole poem is still not feasible. It is evident that Bing translates into English first and then generates the image.
But with some better pre-processing tips, the resulting effect is quite stunning.
Prompt：A withered vine, an old tree, a cawing crow, a small bridge, the flowing water, and a lonely cottage home of a small village, sunset, . Chinese classical painting.
Prompt： A man, a horse, a long narrow path, desolate, rugged, windswept, lonely, melancholy, despondent, sunset, Chinese classical painting.
If the two prompts are combined, the output is:
I think this one is the most appropriate.