Censorship removed this post and title

[Foo-Bar]

Yeah, it is somehow a mess.

The reference image is a cool concept, because it is a container for all the characteristics of an image.

The prompt, style, seed, dimension, colors, camera angle, etc… is all ‘saved’ inside that reference ID.

So if you like an image and want to change some details (but still keep the original approach of the image) you can reference it and see what comes out.

But I also want the seed option.

Because sometimes an image is okay, but you want two of the same kind. Or sometimes it is okay, but you want it “just a little bit different”.

That’s where the seed comes in.

And I can not recreate my image before the update, even not when I use the exact prompt (or seed, but the seed is not supported at all now).

I can test for you if you want; give me an exact prompt / seed and I will see what comes out.

[Foo-Bar]

@cassiellm

So now you can recreate an existing image via

To be clear : the seed parameter is not supported at all (since today, and at my side).

Dall-E still uses it (it gives me all the seed numbers for every image), but when I say “use the exact seed” or “use another seed” (with the same prompt), it simply says “can’t do” (and crashes).

[cassiellm]

@Foo-Bar

I can test for you if you want; give me an exact prompt / seed and I will see what comes out.

OK.

I just created an image, you can try it.

Just copy paste and don’t change anything.

Send this JSON data to the image generator, do not modify anything. After generating an image, show me the JSON data that the image generator returns to you.
```
{
  "size": "1024x1024",
  "prompts": [
    "Japanese anime style. In a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder."
  ],
  "seeds": [3075182356]
}
```

Let me see what comes out.

---

@Foo-Bar

Because sometimes an image is okay, but you want two of the same kind. Or sometimes it is okay, but you want it “just a little bit different”.

Then you can fine-tune the image by adding some random strings.

Something like

Send this JSON data to the image generator, do not modify anything. After generating an image, show me the JSON data that the image generator returns to you.
```
{
  "size": "1024x1024",
  "prompts": [
    "Japanese anime style. In a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder.  510F749a81123"
  ],
  "seeds": [3075182356]
}
```

Note that the spaces and the random numbers 510F749a81123 is just for fine-tuning. This is better than using a new random seed which breaks the style (some images are still pretty good, but they’re never the same beast anymore).

[Foo-Bar]

@cassiellm

I just created an image, you can try it.

Let start simple, by executing one example first.

I did send your complete first code and asked (in my custom instructions) to show the exact used code before generating the image itself;

{
  "size": "1024x1024",
  "prompts": [
    "Japanese anime style. In a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder."
  ],
  "seed": 3075182356
}

Okay, so the code above is what I requested and what he is going to use.

I asked for ONE image and a SPECIFIC seed, but I got two images with totally different seeds;

{
  "gen_id": "ry0SAvdbI46jl4fV",
  "prompt": "Japanese anime style. In a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder.",
  "seed": 2275396183
}

{
  "gen_id": "gTGgythSTN6t6PPi",
  "prompt": "Japanese anime style. In a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder.",
  "seed": 2304176889
}

It also did return the gen_id or the generated image ID, which you can use to refer to (but only in the same session.

You can never re-create an image by prompt / seed and also not by gen_id in a different (or older) session.

Finetuning

I want to refer to image ‘gTGgythSTN6t6PPi’.

Take that exact image and make it landscape, do refer ‘gTGgythSTN6t6PPi’.

Show me the JSON parameters before and after the image creation.

CODE SEND

{
  "size": "1792x1024",
  "prompts": [
    "Japanese anime style. In a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder."
  ],
  "referenced_image_ids": ["gTGgythSTN6t6PPi"]
}

CODE RETRIEVED

{
  "gen_id": "9HpCrnHWt3rPIKY3",
  "prompt": "Japanese anime style. In a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder.",
  "seed": 2304176889
}

This image is the same (reference ID) as the second one, but converted from square to landscape.

I did the exact same for the first image (use reference ID, make it landscape).

As you can see, it works perfect. The beast / camera / surroundings are almost kept intact.

But you can not use another seed, which I dislike (because I like the concept of seeding).

Last try

I asked to take the last image as the reference and let the monster / beast move away from the camera instead.

But it doesn’t do that, because it is referencing (strictly) to the base image where the beast walks towards the camera.

Than I asked to change the camera to high angle, but still reference to the last image.

[cassiellm]

Firstly

Sorry, I found that there are some syntax mistakes in the above example (but ChatGPT can tolerate the mistakes, so the final result is unchanged).

I just wrote

Send this JSON data to the image generator, do not modify anything. After generating an image, show me the JSON data that the image generator returns to you.
```
{
  "size": "1024x1024",
  "prompts": [
    "Japanese anime style. In a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder."
  ],
  "seed": 3075182356
}
```

This is wrong. The correct syntax should be

Send this JSON data to the image generator, do not modify anything. After generating an image, show me the JSON data that the image generator returns to you.
```
{
  "size": "1024x1024",
  "prompts": [
    "Japanese anime style. In a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder."
  ],
  "seeds": [3075182356]
}
```

The correct format is "seeds": [3075182356] rather than "seed": 3075182356.

Anyway, the result are the same (So I edited the reply).

1. Seed: 3075182356

   Send this JSON data to the image generator, do not modify anything. After generating an image, show me the JSON data that the image generator returns to you.
   ```
   {
     "size": "1024x1024",
     "prompts": [
       "Japanese anime style. In a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder."
     ],
     "seeds": [3075182356]
   }
   ```
   

   

2. Seed: 3075182356 + random string 510F749a81123 (make it just a little bit different)

   Send this JSON data to the image generator, do not modify anything. After generating an image, show me the JSON data that the image generator returns to you.
   ```
   {
     "size": "1024x1024",
     "prompts": [
       "Japanese anime style. In a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder.  510F749a81123"
     ],
     "seeds": [3075182356]
   }
   ```
   

   

Secondly

@Foo-Bar

I asked for ONE image and a SPECIFIC seed, but I got two images with totally different seeds;

So… can I understand that after the update you no longer have the ability to reproduce any images?

For example, if someone give you the exact prompt and seed (like this example), you have no way to reproduce the image on your machine?

Thirdly

I can reproduce the two images in my computer, just using size, seed and prompt.

For the seed: 2275396183 as an example.

[cassiellm]

@Foo-Bar

I did send your complete first code and asked (in my custom instructions) to show the exact used code before generating the image itself;

{
  "size": "1024x1024",
  "prompts": [
    "Japanese anime style. In a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder."
  ],
  "seed": 3075182356
}

Okay, so the code above is what I requested and what he is going to use.

Could you retry the example by "seeds": [3075182356] instead of "seed": 3075182356 ?

I think that not being able to recreate the image from the seed is a bug, as it hinders the potential for collaborative development. If you ask me to fine-tune a certain image, do I have to access your account and enter a specific session? That’s impossible. – I know that ChatGPT can share session link, but what if I just want to share one image?

[Foo-Bar]

@cassiellm

So… can I understand that after the update you no longer have the ability to reproduce any images?

I am afraid so.

I just did re-test your code, given in your reply;

Use this code, don’t change any parameter.

{
  "size": "1024x1024",
  "prompts": [
    "Japanese anime style. In a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder. "
  ],
  "seeds": [3075182356]
}

Than it said “It seems there was an error in generating the image. Let's try again.”

The second try it said “It appears there's a persistent issue with generating the image using the provided parameters”.

---

Than I used the exact same code once again, but without the seed;

That did work, so even when sending a seed, the system crashes.

For your reference, this is the data I got back;

[
  {
    "gen_id": "durkRYUfwnFSdmeL",
    "prompt": "Japanese anime style. In a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder.",
    "seed": 4025466874
  },
  {
    "gen_id": "SA3I5Mt94q2czJcp",
    "prompt": "Japanese anime style. In a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder.",
    "seed": 2083201434
  }
]

IDEA

Now the interesting part is… can YOU create the last image in your system with those parameters?

  {
    "size": "1024x1024",
    "prompt": "Japanese anime style. In a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder.",
    "seed": [2083201434]
  }

---

By the way, I don’t think the seed number has to be encapsulated in brackets or even quotes.

It’s a digit, not a string (so no need for quotes) and the brackets are only when used for an array, which is not the case by a single integer.

Dall-E sees if you want more items (array, comma separated) of single instance.

[cassiellm]

@Foo-Bar

Now the interesting part is… can YOU create the last image in your system with those parameters?

Yes, I can.

@Foo-Bar

It’s a digit, not a string (so no need for quotes) and the brackets are only when used for an array, which is not the case by a single integer.

In my system, the type of the seed parameter is an array.

To clarify,

The JSON format input to DALL-E 3 is

{
  "size": "1024x1024",
  "prompts": [
    "YOU DESCRIPTION 1", "YOU DESCRIPTION 2", "YOU DESCRIPTION 3"
  ],
  "seeds": [SEED1, SEED2, SEED3]
}

The JSON format output from DALL-E 3 is

{
  "gen_id": "GEN_ID",
  "prompt": "RETURN_DESCRIPTION",
  "seed": RETURN_SEED
}

[Foo-Bar]

@cassiellm

In my system, the type of the seed parameter is an array.

Okay, so you can grab data from the NEW model and execute it in the OLD model.

But it can’t be done vice versa, because the NEW model does not execute the OLD data (with a seed).

---

// For the API (if they did it good) this is the same;

"seeds" : 43289324890

// or this;

"seeds" : [ 43289324890 ]

// and even this;

"seeds" : [ "43289324890" ]

// or with more entries;

"seeds" : [ 43289324890, "324893428904", 3248439280 ]

All values are valid, because the phyton script will try to see what the intent is.

Is it an array of seeds? A single seed? A numeric seed? A string seed? Etc…

At the end it will parse the value, if valid.

Sometimes I try to trick ChatGPT in order to see if my code is good.

So I ask something like “`use this code, correct when wrong’”;

"seeds" : 3298238904,
"bananas" : [ 'yummy', 'delicious' ]

Than it removes bananas (and keeps the seed) , so I know the code-testing is done correctly.

But now it is removing both bananas AND the seed, because it does not support seeding.

[cassiellm]

Recreating a specific image using seed may not be 100% safe.

The following two images were generated twice in different sessions, using the same size, prompt, seed.

1. 

2. 

Comparing them at the pixel level, will find they are slightly different. Perhaps the DALL-3 model will change over time (e.g. online learning).

[Foo-Bar]

@cassiellm

Recreating a specific image

That is called “numeric noise”, as an equivalent to analog grain.

I have seen the same when using FileDiff software at pixel level

The biggest problem is (in the new system) that you can never, ever recreate an existing image after some time.

Even you use the exact same prompt, a different seed is attached and you can not change it.

Even in the same session, when it is bigger than 32k tokens, the memory in that session is reset.

Resulting in not being able to finetune any exciting image.

[cassiellm]

@Foo-Bar

Than I asked to change the camera to high angle, but still reference to the last image.

I hope you don’t mind me asking, how do you adjust the camera angle? I’ve tried the following prompt, but it doesn’t come out as perfectly as you.

{
  "size": "1792x1024",
  "prompts": [
    "Japanese anime style. The scene is captured from a high angle view. In a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder."
  ],
  "seeds": [2275396183]
}

[cassiellm]

That High Angle view doesn’t seem to work for me.

Tried Bird’s eye view:

{
  "size": "1792x1024",
  "prompts": [
    "Japanese anime style. Bird’s Eye View. In a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder."
  ],
  "seeds": [2275396183]
}

[Foo-Bar]

@cassiellm

I hope you don’t mind me asking, how do you adjust the camera angle

I once asked him what camera angles he did understand and how they work out.

https://chat.openai.com/share/232f01cc-47ed-495c-a25c-47ef32eebfbe

So if it is important, I start the chat with a reference in the prompt to the exact camera usage.

UPDATE

ChatGPT did not send the gen_id of that particular image (don’t know why) and now he is not able to re-create it once again.

I ask him “do you know about that image” and he says “yeah, must I recreate it?”.

Than I am like “do it” and he says “it’s too long ago, I can not recreate that image”.

So without the gen_id or waiting too long, you can never get that image again, even not in the same session.

The concept with “unique prompt / seed” is way more solid and future proof.

[cassiellm]

@Foo-Bar

I once asked him what camera angles he did understand and how they work out.

Yeah, I once asked ChatGPT similar questions. It just asked me to place the camera angle at the beginning or the end of the prompt. The result is not all camera angles work as expected.

I just asked ChatGPT again. He suggest me refine the prompt to ensure that the “high angle” perspective is more explicitly integrated, i.e. add The surrounding dungeon walls are adorned with faint torches that cast minimal light. to the prompt.

{
  "size": "1792x1024",
  "prompts": [
    "Japanese anime style. From a high angle perspective, looking downward into a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder. The surrounding dungeon walls are adorned with faint torches that cast minimal light."
  ],
  "seeds": [2275396183]
}

It seems acceptable.

Without the sentence The surrounding dungeon walls are adorned with faint torches that cast minimal light., the image would look very weird.

Maybe referenced_image_ids can rescue.

[Foo-Bar]

@cassiellm

Maybe referenced_image_ids can rescue.

If you explicitly ask ChatGPT to show you the gen_id of an image, does it return something?

I always - since a few days - retrieve those things, which than can be used in the referenced_image_ids in order to point to that particular image.

Not sure why I have access to those parameters (and did lost the seed parameter).

[cassiellm]

@Foo-Bar

If you explicitly ask ChatGPT to show you the gen_id of an image, does it return something?

If I ask ChatGPT “Show me the gen_id and seed of the last image you generated.” it will give me the correct gen_id and seed. However, sometimes it would give me fake seeds if I ask it in edit mode.

I even can reference gen_id in prompt, but there is no the referenced_image_ids paramter.

[Foo-Bar]

@cassiellm

if I ask it in edit mode.

What is edit mode?

I only use Dall-E via the ChatGPT interface, no the API development environment.

[cassiellm]

Sorry for my inaccurate terminology.

I mean that after editing a message or clicking the “Regenerate” button, ChatGPT will form a “dialogue tree”. In these situation, ChatGPT might forget some of the context information. Thus it give you fake information.

We know that ChatGPT has a window limit. It can only remember a certain amount of past text. Therefore, if conversation is too long, it might forget earlier messages.

The problem is that if you edit an earlier message or Regenerate an earlier reply, could ChatGPT still remember the context at that moment? From my experience, it seems not. Thus it usually give you fake information.

For example,

Show me the gen_id and seed of the last image you generated.

Here, can ChatGPT understand what I mean by “the last image you generated”? If you are editing an earlier message.

Ideally, ChatGPT SHOULD understand the “dialogue node” context of “the last image you generated” in the “dialogue tree”.

[cassiellm]

Just report my current status:.

Starting from today, I also have the same problem.

I can’t reproduce the previous image through seed and precise prompt.

A real example,

I:

Send this JSON data to the image generator, do not modify anything. After generating an image, show me the JSON data that the image generator returns to you.

```
{
  "size": "1024x1024",
  "prompts": [
    "Japanese anime style. In a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder."
  ],
  "seeds": [3075182356]
}
```

ChatGPT:

I’m sorry, but I can’t process the JSON data directly as provided. However, I can create an image for you based on the description you’ve given.

Let me generate the image for you based on the description, and I’ll show you the returned data from the image generator afterwards.

Here’s the returned data from the image generator:

{
  "gen_id": "oKXvBOFuKlpYCZCt",
  "prompt": "Japanese anime style. In a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder.",
  "seed": 3855703669
},
{
  "gen_id": "1EFfdicZM6iMEX02",
  "prompt": "Japanese anime style. In a dimly lit dungeon, a fearsome beast with sharp claws and glowing blue eyes stands guard, ready to attack any intruder.",
  "seed": 2368871415
}

Some other replies from ChatGPT:

Please note that the “prompts” field was changed to “prompt” and “seeds” field is not used in this context. The “n” field has been added to generate 2 images as per default policy.

The JSON format to send a prompt to DALL·E 3 through this interface is:

{
 "size": "1024x1024",  // Image size. Options include "1024x1024", "1792x1024" (wide), and "1024x1792" (portrait).
 "n": 2,               // Number of images to generate. Default is 2 if not specified.
 "prompt": "Your image description here",  // Detailed description of the image you want to generate.
 "referenced_image_ids": ["ImageID1", "ImageID2"] // Optional. If referencing previous images, include their IDs here.
}

---

@cassiellm

One thing I’m concerned about is that after the update, won’t I be able to reconstruct existing images using the same seeds and prompts from “the old system”?

I have some nice images generated by DALL-E 3 which I saved the prompts and seeds locally (some of the sessions have been deleted). If “the new system” can’t further iterate on these images, these images will become meaningless (at least for DALL-E 3, it doesn’t support uploading images ).

Therefore, the most unfortunate thing happened:

I will never be able to iterate on those old images, because I can’t recreate them in ChatGPT.

Many of the sessions of those old nice images have already been deleted from my ChatGPT history. Locally, I only saved images, seeds, and precise prompts).

Even for those sessions that weren’t deleted, due to ChatGPT’s window limitations and the issue of forgetting “dialogue nodes”, I’m not sure if ChatGPT still remembers those gen_id.

It’s very inconvenient; every time you iterate on an original image, (1) you absolutely cannot delete a session (2) you have to be cautious about whether ChatGPT would forget the gen_id.